Research Article – PLOS Currents Outbreaks http://currents.plos.org/outbreaks Wed, 07 Nov 2018 23:18:12 +0000 en-US hourly 1 https://wordpress.org/?v=4.5.3 Real Time Forecasting of Measles Using Generation-dependent Mathematical Model in Japan, 2018 http://currents.plos.org/outbreaks/article/real-time-forecasting-of-measles-using-generation-dependent-mathematical-model-in-japan-2018/ http://currents.plos.org/outbreaks/article/real-time-forecasting-of-measles-using-generation-dependent-mathematical-model-in-japan-2018/#respond Mon, 15 Oct 2018 16:45:43 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=82876 Background: Japan experienced a multi-generation outbreak of measles from March to May, 2018. The present study aimed to capture the transmission dynamics of measles by employing a simple mathematical model, and also forecast the future incidence of cases.

Methods: Epidemiological data that consist of the date of illness onset and the date of laboratory confirmation were analysed. A functional model that captures the generation-dependent growth patterns of cases was employed, while accounting for the time delay from illness onset to diagnosis.

Results: As long as the number of generations is correctly captured, the model yielded a valid forecast of measles cases, explicitly addressing the reporting delay. Except for the first generation, the effective reproduction number was estimated by generation, assisting evaluation of public health control programs.

Conclusions: The variance of the generation time is relatively limited compared with the mean for measles, and thus, the proposed model was able to identify the generation-dependent dynamics accurately during the early phase of the epidemic. Model comparison indicated the most likely number of generations, allowing us to assess how effective public health interventions would successfully prevent the secondary transmission.

]]>

Introduction

Measles is a highly contagious viral infectious disease transmitted by aerosol 1. Clinical symptoms of measles include fever, sore throat, conjunctivitis, and rash, and it can potentially be lethal to infants and children, leading to serious complications including encephalitis and neurological complications 2. In Japan, since the year 1978, routine vaccination against measles have started, and a two-dose regimen has been introduced among birth-cohorts born in and after 1990, contributing to reducing the burden of measles by elevating the immunity level in the population 3,4. The transmission of the virus in Japan has not been sustained, and the Measles Regional Verification Commission of the World Health Organization (WHO) Regional Office for the Western Pacific verified Japan as having achieved measles elimination in March 2015 5. However, global circulation of the virus continues to pose a risk of sporadic outbreaks to Japan 6.

From March 2018, an abrupt outbreak in Okinawa has been notified ahead of the “Golden Week”, the longest vacation period of the year (i.e., from 28 April to 6 May 2018). The index case was a 30-year-old Taiwanese man who had a travel history to Thailand in early March. On 17 March 2018, he flew to Okinawa, and on the third day of his stay in Okinawa, he sought for medical service. Following an incubation period of 11-12 days after the diagnosis of the index case 7, multiple generations of local cases were identified in Okinawa prefecture. Cases originating from Okinawa prefecture produced multiple chains of transmission, bringing a total number of confirmed cases to 124. The spread could not have been contained within Okinawa, and spread to Aichi prefecture, Kanagawa prefecture, and Tokyo Metropolis.

During this outbreak, measles cases have been confirmed at governmental diagnostic research facilities and reported in real-time. Each report was regarded as a snapshot of the growing epidemic curve that was used for forecasting of the future course of the outbreak. To understand better the transmission dynamics during the course of an outbreak, we implemented the future forecast to infer public health control activities. While not explicitly assessing the control activities in our exposition, the purpose of the present study is to capture the transmission dynamics of measles by employing a simple parsimonious mathematical model and to forecast future generations of measles incidence.

Methods

Epidemiological data

Measles is clinically diagnosed by the presence of a generalized rash, fever, and catarrh symptoms, such as cough, coryza, or conjunctivitis, and then laboratory confirmed. The laboratory confirmation is performed by detection of measles-specific immunoglobulin M (IgM) antibodies 8 or real-time reverse transcription polymerase chain reaction (rRT-PCR). There is an additional clinical form of measles, “modified measles”, that usually exhibits only one of three symptoms, is laboratory confirmed, and has a milder clinical course of illness. The present study rests on governmental reports based on outbreak investigation in Japan 9, including local governmental reports from prefectures with at least one case, i.e., Okinawa, Aichi, Kanagawa and Tokyo Metropolis 10,11,12,13. We retrospectively scanned all real-time reports of the outbreak and reconstructed the epidemiological dynamics of measles that developed from the identical case. Dates of illness onset and laboratory confirmation, retrieved from those governmental reports, allowed us to characterize the epidemic dynamically evolving in time.

Inference procedure

Due to close contact tracing practice, we assumed that all cases were certainly diagnosed and reported. To quantify the underlying epidemiological dynamics and delay distribution from illness onset to laboratory confirmation, we employed the maximum likelihood estimation technique. Specifically, we considered the total (composite) likelihood function LΣ consisting of two parts, each corresponding to different pieces of the dataset, i.e., (i) individual datasets of the dates of illness onset and laboratory confirmation, and (ii) the number of new cases by the date of illness onset. The first part allowed us to identify the distribution of the time delay that was subsequently used to predict the number of cases who have yet to be confirmed. We introduced a probability density function h(dn; θh) that measured the probability of the case n = {1 … N} to be confirmed on dn days after the onset of symptoms. We assumed that the distribution h does not vary as a function of calendar time during the course of the outbreak. An alternative formulation of a time-varying distribution h did not improve the model fit and hence was discarded here (see Appendix A). Using the first part of the dataset, we arrived at the likelihood function of the form:

We assumed that h follows a Weibull distribution with parameters the mean and variance vh (θh = {μh, vh}). The second part of the likelihood utilizes a part (ii) of the dataset that describes the incidence on a day t = {1 … T} denoted as it. The incidence follows a Poisson distribution:

where H is a cumulative distribution of the delay h, counted backwards in time from the day of the update publication T.

The incidence function Λt(θΛ) is modeled by sequential generation process (hereafter, referred to as the generation-dependent model). Each new case has an ability to generate new secondary infections with the probability density function of the generation time gt. In brief, we describe this by the following epidemiologic process. The index case solely belongs to the first generation. It first generates new R1 cases distributed in time according to the distribution gt that constitute the second generation. Each secondary case subsequently generates R2 tertiary cases according to the same distribution gt resulting in the third generation. Because of independently and identically occurring transmission events, the total number of cases at calendar time t is given by the formula: R1(gt + R2(gg)t), where the symbol “∗” stands for the convolution operator of two functions on its left- and right-hand sides. Specifically, a convolution operator of the functions f1 and f2 at time t yields the following formula:

where f2(0) is equal zero. The above mentioned method is restricted to three generations, however, we can assume any arbitrary number of new generations for describing an epidemic curve. For example, if we account for up to four generations, the total number of cases is written accordingly as: R1(gt + R2((gg)t + R3(ggg)t)).

However, the rate Λt needs to be normalized to the expected cumulative number of all cases K, which means:

where Rm is the effective reproduction number of the (m + 1)-th generation, NR represents a normalization constant restricting the total number of cases to K. Hence, NR = R1 + R1R2 + R1R2R3. See Appendix B for the derivation of generation-based model. Due to the normalization, the parameter R1 cannot be recovered, only its lower bound can be identified by manually counting the number of secondary cases who can certainly identified as caused by the index case. The model fit for shorter time horizon of forecasting may require a smaller number of generations, and thus, the latest formula would need to be modified (e.g. R3 = 0 in case of three generations only, and R2 = R3 = 0 in case of two generations only, governing the entire dynamics of the observed epidemic data). As for the generation time distribution gt, we employ a gamma distribution with the mean 11.7 days and variance 9.0 day2 (the average of two previously reported estimates 7).

The total likelihood is:

subject to maximization with respect to five parameters (θ = {K, R2, R3, μh, vh}). By using an equivalent minimization of the negative logarithm of the likelihood, we gain the optimal parameter values θ = θ0 as well as the Hessian matrix H(θ0). To reconstruct the confidence intervals and the compute the prediction interval, we implement the matrix H into the parametric bootstrapping. First, we design a dataset that consists of model parameters sampled from the normal distribution with the mean θ0, and standard deviation σ equal to the square root of diagonal elements of the inverse Hessian matrix (σ2 = diag(H-1(θ0))). Then for each identical set of parameters, we obtain a possible variation in estimated parameter values. Finally, by taking 2.5th and 97.5th percentile points of the simulated distributions, we obtain 95% prediction intervals of incidence function.

Forecasting procedure

To perform forecasting exercise, we used an epidemic curve of new measles cases, routinely collected and updated every eight days. As a result, we obtained multiple snapshots of the epidemic curve, all initiated with the date of exposure to the index case on 17 March 2018, but constrained by the date of publication (ranged from 1 April to 25 May with a time step of eight days). Data points of each epidemic curve were then imputed to our model to identify expected number of cases over the time interval of that epidemic curve. Furthermore, the cases were forecasted for an extended time period, up until 8 June.

Results

The number of new cases of measles by the date of illness of onset and date of laboratory conformation are shown as Figures 1A and 1B, respectively. As of 21 August 2018, a total of 124 laboratory confirmed cases have been reported in Japan, of which 99 cases have been in Okinawa, 23 in Aichi, 1 in Kanagawa prefecture, and 1 in Tokyo.

(A) Date of illness onset of measles cases reported in Tokyo Metropolis, Kanagawa, Aichi, and Okinawa prefecture, Japan. Illness onset was unknown for 6 cases notified in Okinawa prefecture, thus, was assumed to be 5 days before laboratory confirmation. (B) Date of laboratory confirmation of measles cases reported in Tokyo Metropolis, Kanagawa, Aichi, and Okinawa prefecture.

Fig. 1: Date of illness onset and laboratory confirmation of reported measles cases in Japan, March-May, 2018.

(A) Date of illness onset of measles cases reported in Okinawa, Aichi, Kanagawa prefectures, and Tokyo Metropolis, Japan. Illness onset was unknown for 6 cases notified in Okinawa prefecture, thus, was assumed to be 5 days before laboratory confirmation. (B) Date of laboratory confirmation of measles cases reported in Okinawa, Aichi, Kanagawa prefectures, and Tokyo Metropolis.

Using the observed epidemiological data from Okinawa, Aichi, Kanagawa, and Tokyo reported from 1 April to 25 May 2018, unknown parameters were estimated as shown in Figure 2. In addition, a penalized likelihood for the models of different number of generations was compared based on Akaike Information Criterion (AIC). The minimal value of AIC was used to determine the best-fit number of generations for a given date of publication of the dataset.

The course of outbreak observed before 9 April fitted well using only two generations. Afterwards, the third generation was identified, and the fourth generation appeared since 25 April. The effective reproduction number of the third generation R2 became greater than one on 9 April. When a new generation was identified to better explain the observed incidence pattern, the model with greater number of generations fitted better than the model with fewer generations. Importantly, our model explicitly accounted for the time delay from illness onset to diagnosis, and thus the effective reproduction number of the most recent generation avoided serious underestimation. However, the expected value of R4 during the early stage was smaller than during the later stage – the number of cases in the fifth generation was not substantial in May, and thus R4 was accompanied by a wider confidence interval.

Table_1

Fig. 2: Estimated parameters values and model comparison by the epidemic date of forecasting.

“#” denotes the assumed number of generations in the model. ht is the probability mass function from the time of illness onset to laboratory confirmation. Rm is the reproduction number of (m + 1)-th generation. AIC is Akaike information criterion. RMSE is the root-mean-square error. K is the estimated total number of symptomatic cases in Japan. Selected models with minimal AIC are shown in red. 95% confidence intervals (CIs) for each model parameter are shown in brackets.

Using the latest snapshot of the epidemic curve published on 25 May, the mean delay from illness onset to confirmation was 4.5 days (95% CIs: 4.0-5.0), and the variance was 6.1 day2 (95% CIs: 4.3-9.0). The total number of symptomatic cases K which was unknown on the date of publication as some cases with symptoms could still undertake laboratory identification, was estimated as 123.2 (95% CIs: 102.1-145.4). The obtained estimate was close to the observed total number of 123 cases, excluding the index case.

Figure 3 shows the forecasted course of the measles epidemic by using the proposed generation-dependent mathematical model, and the data of confirmed cases of each epidemic curve according to different confirmed date. In the first stage, the model describes only the initial part of the outbreak, but the estimates become certainly improved and the 95% prediction intervals progressively become narrower as more data are used.

Performance of forecasting for each epicurve (Legend) is compared to the number of reported cases in the latest update (bar chart in grey). Lines in dashed denote the forecasting part for each snapshot of the epicurve.

Fig. 3: Real time forecasting result of measles in Japan, 2018.

Performance of forecasting for each epicurve (legend) is compared to the number of reported cases in the latest update (bar chart in grey) by date of illness onset of measles cases (A) and date of laboratory confirmation of measles cases (B). Dashed lines denote the forecasting part for each snapshot of the epicurve.

The following Video available online (Figure 4, doi:10.6084/m9.figshare.6991367) presents an extended version of Figure 3 with daily snapshots of the epicurves. As we see, there is a greater degree of uncertainty in the future forecast once a new generation of cases appears and is accounted in the model.

Forecast_Measles2018

Fig. 4: Animated real time forecasting result of measles in Japan, 2018, with daily snapshots of epicurves.

Discussion

The present study tackled real-time forecasting of measles, employing a generation-specific modelling approach. A simple functional model with generation structure was employed, and the time delay from illness onset to diagnosis was explicitly taken into account. The proposed model helped not only to forecast the future incidence but also to obtain the generation-specific estimates of the effective reproduction number. AIC values helped to identify the most likely number of generations in real-time, allowing us to assess how good public health interventions successfully prevented transmission events during the outbreak. To our knowledge, the present study is the first study to apply the functional generation-dependent model to the context of real-time forecasting.

There are two take home messages. First, the generation-dependent mathematical model successfully helped to anticipate the likely size of the future epidemic in real time. Because the variance of the generation time for measles is relatively limited compared to the mean, the generation-specific number of cases was even manually identified during the early phase of the outbreak 14. This was consequently used in the model. Nevertheless, the reliance on the number of generations can also be regarded as a disadvantage – the model is unable to forecast future generations without appearance of a cluster of likely new cases from the next generation in empirical data or without imposing strong assumptions, e.g., that the effective reproduction number remains the same for a series of the next generations. Thus, we may regard our model as yielding the real-time forecast only for a minimal bound of the future incidence.

Second, the estimation of the effective reproduction number as the weight for the mixture distribution of the generation time is also a side-product of the model (see Appendix B). Without doubt, the reproduction number helps to evaluate preventive measures during the outbreak. Nevertheless, our study also addressed a possible underestimation of the effective reproduction number for the latest generation once considering an explicit time delay from illness onset to laboratory confirmation. Although we did not incorporate stochasticity in the functional model, our model was able to capture the mechanistic pattern of the transmission dynamics.

Few technical limitations must be described. First, the absence of stochasticity in the transmission process is a systematic limitation of the proposed model. To capture the stochasticity of the transmission process, we must employ a stochastic process model to describe the transmission event, e.g., a branching process or a renewal process. Second, we did not explicitly use susceptibility of the exposed population, and also the background information on the traced contacts. While those datasets were not routinely collected, their use could help increase the validity of the forecast. Third, vaccination history of cases was not taken into consideration. Depending on residual immunity, we may observe a different clinical form of measles, i.e., modified measles. This could lead to a different (potentially longer) time delay from illness onset to diagnosis compared with the primary form of measles. Lastly, our assumptions included a fixed delay distribution function over the whole period of the outbreak. As we additionally verified, the inclusion of a step-like temporal dependence of the mean and variance of the delay function with given switching times (e.g. 29 March and/or 3 April as the dates of raised awareness 15) did not improve the model fit.

In conclusion, we demonstrated a simple generation-dependent model that was able to adequately capture an observed transmission pattern of the measles outbreak in Japan, 2018. The proposed model also helped predict the future incidence and evaluate public health control measures. Polishing the forecasting model further, we can achieve an eventual routine forecast and evaluation the outbreaks while maintaining the model structure as simple as possible.

Competing Interests

The authors declare no competing interests.

Data Availability

The code snippets used for simulations and generation of figures as well as the epidemiological count data are accessible from the GitHub repository: https://github.com/aakhmetz/MeaslesJapan2018.

Corresponding Author

Hiroshi Nishiura (nishiurah@med.hokudai.ac.jp)

Appendix

A. Time-varying delay function

Here we describe the fitting procedure when the delay distribution function h is a mixture of two distributions. The first distribution describes all cases whose times of illness onset are prior to a calendar time τ. In our exposition, the delay function h follows a Weibull distribution with parameters θh(0) = {μh(0), νh(0)}. The second distribution describes all cases with the time of illness onset later than a calendar time τ. It also follows a Weibull distribution with a set of parameters θh(1) = {μh(1), νh(1)}. The likelihood to describe the time delay from illness onset to laboratory confirmation is given by the formula:

where tn is the time of illness onset for each particular case: n = {1 … N}; I(x) is a step function: it equals to one when its argument x is non-negative, and zero otherwise. Analogously, we characterize the likelihood for new measles cases by the formula:

Whereas, the total (composite) likelihood is given by a product of two likelihoods written above:

The total likelihood is maximized with respect to each parameter in the set θ, consisting of (4 + m) parameters (m is the number of generations). Hence, the penalized likelihood used for model comparison based on AIC values can be defined as: 2(4 + m – ln LΣ(θ; dn, tn, it)).

Model performance is shown in Figure 5 that can be compared with previous case of time-independent distribution h shown in Figure 2.

Table_2

Fig. 5: Estimated parameter values and model comparison for a simple case of time-varied distribution of the delay h.

For any epicurve only the cases with minimal AIC values over a set of varied number of generations are shown. The switch in delay function indicates the optimal switching time, i.e., the calendar time on which the distribution is considered to have changed. The mean and variance of the delay distribution function before the switching day are indicated by the variable ht(0), after the switching day by the variable ht(1). The AIC values for a model with fixed distribution of the delay are shown in the last column, while the minimal AIC values are additionally indicated in red.

B. Derivation of the generation-based model

Our generation-dependent model rests on a well-known renewal equation, i.e.,

where R(t) represents the instantaneous reproduction number at calendar time t, i(t) is the incidence, and g(s) is the probability density function of the generation time of length s. For any t > 0, the density of incidence at time t is given by the generation expansion 16:

where im(t) results from the iteration process:

Here Rm is the cohort-reproduction number of generation m (or “(m + 1)-th” generation in the main text if we included the index case as generation 1). The integral of im(t) over t gives the total size of generation m, and thus, Rm can be mathematically interpreted as the asymptotic per-generation growth factor of the genealogy, consistent with the definition of the basic reproduction number 17.

Replacing the right-hand side of (B2) by that of (B3), we obtain Λt in the main text. As such, it should be noted that we perform forecasting by estimating the generation-dependent average number of secondary cases generated by a single primary case, which is interpreted as the cohort reproduction number (i.e., the average number of secondary cases generated by a primary case who was born at calendar time t), and not as the instantaneous reproduction number.

]]>
http://currents.plos.org/outbreaks/article/real-time-forecasting-of-measles-using-generation-dependent-mathematical-model-in-japan-2018/feed/ 0
Measuring Mosquito-borne Viral Suitability in Myanmar and Implications for Local Zika Virus Transmission http://currents.plos.org/outbreaks/article/measuring-mosquito-borne-viral-suitability-in-myanmar-and-implications-for-local-zika-virus-transmission/ http://currents.plos.org/outbreaks/article/measuring-mosquito-borne-viral-suitability-in-myanmar-and-implications-for-local-zika-virus-transmission/#respond Fri, 28 Sep 2018 09:37:33 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=81191 Introduction: In South East Asia, mosquito-borne viruses (MBVs) have long been a cause of high disease burden and significant economic costs. While in some SEA countries the epidemiology of MBVs is spatio-temporally well characterised and understood, in others such as Myanmar our understanding is largely incomplete. 

Materials and Methods: Here, we use a simple mathematical approach to estimate a climate-driven suitability index aiming to better characterise the intrinsic, spatio-temporal potential of MBVs in Myanmar. 

Results: Results show that the timing and amplitude of the natural oscillations of our suitability index are highly informative for the temporal patterns of DENV case counts at the country level, and a mosquito-abundance measure at a city level. When projected at fine spatial scales, the suitability index suggests that the time period of highest MBV transmission potential is between June and October independently of geographical location. Higher potential is nonetheless found along the middle axis of the country and in particular in the southern corridor of international borders with Thailand. 

Discussion: This research complements and expands our current understanding of MBV transmission potential in Myanmar, by identifying key spatial heterogeneities and temporal windows of importance for surveillance and control. We discuss our findings in the context of Zika virus given its recent worldwide emergence, public health impact, and current lack of information on its epidemiology and transmission potential in Myanmar. The proposed suitability index here demonstrated is applicable to other regions of the world for which surveillance data is missing, either due to lack of resources or absence of an MBV of interest.

]]>

Notice of Correction

10 October 2018: PLOS Currents – Correction: Measuring Mosquito-borne Viral Suitability in Myanmar and Implications for Local Zika Virus Transmission. PLOS Currents Outbreaks. 2018 Oct 10 . Edition 1. doi: 10.1371/currents.outbreaks.9934c8779f27f8fa6e4d59d3197dff85. View Correction.

Introduction

Common mosquito-borne viruses (MBVs) of global health concern include the dengue (DENV), chikungunya (CHIKV), Zika (ZIKV), yellow fever (YFV), Rift Valley fever (RVFV), West-Nile (WNV) and Japanese encephalitis (JEV) viruses. Due to ongoing globalization and climatic trends that favour the establishment of vectors and movement of infectious hosts, these pathogens are becoming increasingly detrimental for human public health 1,2,3.

The evolutionary and host-pathogen history of MBVs is vastly diverse, but the population biology of these viruses shares one unifying characteristic: their epidemiological dynamics and epidemic behaviour is inherently linked to the underlying population dynamics of their vector-species. Mosquito-population dynamics are known to be dictated by a wide range of factors, such as climate, altitude, population density of humans or other animals, air or waste pollution levels, and natural or artificial water reservoirs 4,5,6,7. While most of these factors can dictate absolute population sizes (carrying capacity), seasonal oscillations are largely driven by natural climate variations.⁠ South East Asia (SEA) is the most densely populated region of the world and has experienced rapid urbanization in the past century. Such demographic factors and tropical climate are believed be the main drivers of the success of Aedes mosquitoes in the region, the main genus for the transmission of MBVs such as DENV, CHIKV and ZIKV. Historically, surveillance of MBVs has been highly heterogeneous in the region. For instance, a few countries such as Vietnam, Thailand and Singapore often serve as global references for DENV epidemiology, reporting spatio-temporal epidemiological and genetic data spanning several decades 8,9. Other countries, such as Myanmar, report epidemiological data at lower spatio-temporal resolutions 10,11,12. For instance, published DENV case counts over time are reported at national level with total counts per year at the state level, but spatio-temporal data at the district or lower levels is not freely available. In the context of countries with incomplete data coverage for a robust understanding of the local epidemiological determinants and epidemic potential of MBVs, mathematical frameworks, and in particular dynamic models, are essential to close the gaps in knowledge.

Recently, we have developed and applied a climate-driven mathematical framework to study epidemics of MBVs such as DENV 13,14 and ZIKV 15⁠. The success of this framework stems from its data-driven approach, including temporal climatic series used to parameterise both vector and viral variables under mathematical relationships derived from experimental studies. The reliance on local climatic variables as its main input makes the framework general enough to be applied to different geo-locations. Here we translate our experience with this framework to introduce an MBV suitability index, which we apply to the context of Myanmar. With it, we estimate spatio-temporal patterns of suitability across the country, validating it against existing but incomplete data. Our results contribute to a better understanding of Myanmar’s MBV transmission potential in both time and space. We discuss the public health and control implications of our findings both generally for MBVs and in particular for ZIKV.

Methods

Our approach develops from a climate-driven, mosquito-borne mathematical model of viral transmission that has been successfully applied to three MBV epidemics: for the 2012 dengue serotype 1 outbreak in the island of Madeira (Portugal) 13⁠, the 2014 dengue serotype 4 outbreak in Rio de Janeiro (Brazil) 14 and the 2015-2017 ZIKV outbreak in Feira de Santana (Brazil) 15⁠. In these case studies, given the availability of reported epidemic curves at appropriate spatio-temporal scales, deterministic simulations were used with a Markov chain Monte Carlo fitting approach to derive key local eco-epidemiological parameters, allowing of the estimation of the basic reproductive number (R0) and effective reproductive number (Re).

In the context of Myanmar and the lack of reported case counts at the subnational level, we addressed the potential for MBVs transmission by focusing on the model’s equation for R0 15, a similar starting point of a recent and successful strategy implemented in the context of YFV in the African continent 16⁠. R0 is the sum of the transmission potential of each adult female mosquito per human, across the total number of female mosquitoes per human (M), in a totally susceptible human population. Hence, R0 can be expressed as the product of M with each individual mosquito transmission potential P(u,t):

The R0 expression is dependent on time-varying humidity (u) and temperature (t); M is the number of adult female mosquitoes per human (V/N, with N the human population size and V the adult female mosquito population size); av(u) is the mosquito biting rate, dependent on humidity; Φv-h(t) is the probability of transmission of the virus by infected-mosquito to human, per bite, dependent on temperature; Φh-v is the probability of transmission from infected-human to mosquito, per bite; 1/γv(t) is the extrinsic incubation period, dependent on temperature; 1/γh is the intrinsic incubation period; 1/μv(u,t) is the mosquito life-span, dependent on humidity and temperature; 1/σh is the human infectious period; and 1/μh is the human life-span. Each of the climate dependent functions was previously determined by laboratory estimates of entomological data (equations 20-37, 15).

The individual-mosquito transmission potential P(u,t) is a positive number, with P>1 indicating the capacity of a single adult female mosquito to contribute to epidemic expansion, and P>1 indicating otherwise. However, this threshold around 1 does not necessarily equate to the classic epidemic threshold of R0>1 (or Re>1), since P(u,t) critically ignores the total number of female mosquitoes per human (M). In other words, an epidemic threshold may be reached, for instance, with P>1 if M>>1. Here, we argue and demonstrate that P(u,t) holds critical information on transmission seasonality (timing) and amplitude (relative epidemic potential between seasons and regions), both driven by the inherent climatic variables affecting viral and entomological factors. We denote P(u,t) as the mosquito-viral suitability index P, using a complementary terminology to existing vector suitability indices which more generally consider entomological factors and / or vector-population sizes 17,18,19. As seen above, P(u,t) is a complex expression containing human, entomological and viral factors, and can thus be parametrized for any species of virus, host or vector, and estimated for any region for which humidity and temperature are available.

While some parameters used to calculate P can be quantified through known constant values, others follow mathematical expressions that depend on three scaling factors, α, ρ and η, that are used to modulate the baseline relationships between climatic variables and entomological parameters:

Here, a is the baseline biting rate, and the terms marked with * are the actual functions defined in the empirical studies, which can be found in Materials and Methods of Lourenco et al. 15. The multiplicative coefficients η and α are used in the temperature dependent components of the adult mosquito mortality and incubation period, respectively. Their inclusion does not alter the relative effect of temperature variation on the entomological parameters per se, but allows for the parameter’s baselines to be different from the ideal laboratory conditions of the original research (e.g. 20). In practice, the effect of temperature on these parameters can be considered to be the same as observed under laboratory conditions if η~1 and α~1, or weaker if >1 or higher if >1. The exponential parameter ρ allows instead to modulate the strength by which adult mosquito mortality and biting rate react to deviations from local mean humidity. In practice, the effect of humidity can be switched off when ρ tends to 0 and made stronger when ρ >1. For a discussion on possible biological factors that may justify these factors divergence from 1 please refer to the original description of the method 13 and in a separate study by Brady and colleagues 21.

Fitting exercises to MBV epidemic curves that would allow for quantitative estimations of scaling factors α, ρ and η were not possible for Myanmar due to the lack of reported cases at appropriate spatio-temporal scales. To overcome this limitation, we ran a parameter sweep on the three factors in the range 0-10, and drew the combination of three values that would derive a yearly mean life-span of adult mosquitoes of ~9 days and an extrinsic incubation period of ~5 days. These heuristics are based on prior knowledge for ZIKV and DENV transmission estimations with the same model in three different regions 13,14,15, which are themselves informed by reported biological ranges for Aedes mosquitoes 15,21,22,23,24 . Please refer to the Data section for a description on the epidemiological and climate time series used in this study and the section Parameters specific to Myanmar section for all subnational values found and used for the scaling factors α, ρ and η. Constant parameter values used were: human lifespan of 64 years 25, human infectious period of 5.8 days, biting rate of 0.25 per day, human incubation period of 5.9 days and the infected-human to mosquito probability of transmission per bite of 0.5 (as previously modelled 15).

Results

We first tested the index P in the context of publicly available DENV case count data at the national-level for the period between 1996 and 2001 12. For this, we estimated P for each district using available local climatic data (2015-2016), further aggregating and averaging P across all districts of the country and per month (Figures 1 A1-2). While the epidemiological and climatic data available for the analysis were from different time periods, we found that the estimated P for both 2015 and 2016 presented seasonal fluctuations in sync with mean DENV counts from multiple years (1996-2001). The dynamics of P at the country level further presented key signatures in accordance to Myanmar’s climatic seasons. Namely, (i) a sharp increase in transmission potential during May and June, coincident with the onset of the rainy season (Jun – Oct), and (ii) a trough in potential in the middle of the hot and dry season (Mar – May). A linear correlation between mean DENV counts (1996-2001) and mean index P (2015-2016) showed that ~76% of the variation in case count dynamics could be explained by the index with statistical significance (p-value=2.4×10-4, Figure 1 B3).

Panel A1 presents the mean estimated index P across Myanmar for 2015 (red dots, locally weighted smoothing bounds within red area) superimposed on monthly case counts of DENV for several transmission seasons (1996-2001, blue lines) in Myanmar. The black line is the mean DENV case counts 1996-2001. Panel A2 is the same as B1 but with P estimated for 2016. Panel A3 is a linear regression of mean DENV case counts (1996-2001) versus the mean index P (2015-2016) as displayed in panels A1-2. Panel B1 presents the mean estimated index P in Yangon for 2015 (red dots, locally weighted smoothing bounds within red area) superimposed on monthly number of major breeding containers in 2011 (green lines) for Yangon. Panel B2 is the same as B1 but with P estimated for 2016. Panel B3 is a linear regression of major breeding sites (2011) versus the mean index P (2015-2016) as displayed in panels B1-2.

Fig. 1: MBV suitability index P and ento-epidemiological time series in Myanmar.

Panel A1 presents the mean estimated index P across Myanmar for 2015 (red dots, locally weighted smoothing bounds within red area) superimposed on monthly case counts of DENV for several transmission seasons (1996-2001, blue lines) in Myanmar. The black line is the mean DENV case counts 1996-2001. Panel A2 is the same as B1 but with P estimated for 2016. Panel A3 is a linear regression of mean DENV case counts (1996-2001) versus the mean index P (2015-2016) as displayed in panels A1-2. Panel B1 presents the mean estimated index P in Yangon for 2015 (red dots, locally weighted smoothing bounds within red area) superimposed on monthly number of major breeding containers in 2011 (green lines) for Yangon. Panel B2 is the same as B1 but with P estimated for 2016. Panel B3 is a linear regression of major breeding sites (2011) versus the mean index P (2015-2016) as displayed in panels B1-2.

Because we propose the index P as a measure of suitability independent of the total number of (female) mosquitoes, it is critical to have empirical support on whether its oscillatory behaviour (timing and amplitude) correlates with local measures of mosquito population size. Studies including mosquito surveys in Myanmar are scarce in the literature, mostly targeting vectors related to malaria transmission and focusing on their spatial distribution and species diversity (e.g. 26,27,28). Here, we use data collected in a study of Aedes aegypti abundance in Yangon for the year of 2011 29. As far as we know, this is the only study with larval indices measured monthly over the period of at least one year. We compared the published data with our estimations of index P for Yangon in 2015 and 2016 (Figures 1 B1-2). As previously shown for DENV counts (Figure 1 A1-2), the seasonal fluctuations of index P and larval indices were synchronous in time. A linear correlation between Yangon’s larval indices (2011) and index P (2015) showed that ~76% of the variation in mosquito abundance measure, a proxy for adult mosquito population size, could be explained by the index P with statistical significance (p-value=2.2×10-4, Figure 1 C3).

As in other places of the world presenting endemicity for MBVs, Myanmar is likely to present significant heterogeneities between districts (and states) in terms of suitability (transmission potential). Apart from studies reporting differences in the total number of MBV case counts across the country, an assessment of suitability in space has not been previously done for Myanmar. Since our results suggested that the index P contains information on the timing and amplitude of observed DENV counts and mosquito population size (Figure 1), we explored the spatial variation of index P across the country.

We looked at the spatial variation of the index P (2015-2016), focusing on its average within the cool dry (Nov – Feb), hot dry (Mar – May) and wet seasons (Jun – Oct) (Figure 2A), and found significant differences between seasons. Namely, the cool dry season presented a generally lower index P across space, at a time when climatic conditions are expected to be less favorable for the mosquito and therefore for suitability. In contrast, the wet season presented the highest index P across space. Importantly, the latter occurs in the same time period (Jun – Oct) in which DENV counts and mosquito abundance measures also peak. Significant variation within each season was also observed, with the cool and dry season presenting more homogeneous suitability across space (standard deviation, SD=3.86) and the wet season presenting the most heterogeneous suitability (SD=4.17).

We next attempted to correlate the spatial distribution of suitability with local mosquito abundance measures and MBV case counts. However, we found no data with spatial resolution for abundance, and found only DENV counts at the level of Myanmar’s states for the year 2015 as reported by the Ministry of Health and Sports 30. Similarly to the approach applied for districts, we calculated the yearly average index P per state using climate data for both 2015 and 2016 (Figure 2B) and compared its spatial distribution with DENV counts per state for 2015 (Figure 2C). States presenting higher average suitability were located across the centre of the country, but particularly in the south, sharing a border with Thailand. The distribution of DENV counts presented a generally similar spatial signature co-localised in the centre and south of the country (Figure 2C). The Pearson’s correlation between mean yearly suitability and yearly DENV counts of the state-based two maps was 0.56 (p-value=0.033).

Panel A shows maps of Myanmar coloured according to mean index P per district in different seasons of the year (as labelled in each map). Panel B presents the yearly mean index P per state in 2015 with borders of neighboring countries (named) shown in light blue. Panel C presents the number of DENV cases per state in 2015. Panel D presents a sensitivity exercise showing the critical index P (~1.5) for which the spatial distributions of dengue cases and mean index P are most correlated in 2015. Colored lines show the amount of time (T) each state spends with index P above a certain threshold (colors related to 2015 DENV case counts, as in map C). Points present Pearson’s correlation coefficient between T and dengue case counts with significant correlations in blue. Dashed vertical lines signal the T values for which the minimum (no) correlation is found. In all panels: all model parameters as described in Methods section, except for α, ρ and η as described in section Parameters specific to Myanmar.

Fig. 2: MBV suitability index P and spatial distribution of epidemiological data in Myanmar.

Panel A shows maps of Myanmar coloured according to mean index P per district in different seasons of the year (as labelled in each map). Panel B presents the yearly mean index P per state in 2015 with borders of neighboring countries (named) shown in light blue. Panel C presents the number of DENV cases per state in 2015. Panel D presents a sensitivity exercise showing the critical index P (~1.5) for which the spatial distributions of dengue cases and mean index P are most correlated in 2015. Colored lines show the amount of time (T) each state spends with index P above a certain threshold (colors related to 2015 DENV case counts, as in map C). Points present Pearson’s correlation coefficient between T and dengue case counts with significant correlations in blue. Dashed vertical lines signal the T values for which the minimum (no) correlation is found. In all panels: all model parameters as described in Methods section, except for α, ρ and η as described in section Parameters specific to Myanmar.

Again, since we propose the index P as a measure of suitability independent of the number of (female) mosquitoes per human (M), a typical threshold of P=1 may not be adequate to speculate on local epidemic potential, given that an R0>1 can be achieved with P>1 and M>1 (although the results from Figure 1 suggest that both the timing and amplitude of P are highly informative). We hypothesized that the time spent above a certain value (threshold) of suitability could be a better proxy for local, yearly epidemic potential. A sensitivity analysis was performed for 2015, the only year for which we had both climate input (index P) and DENV counts, with the intent of searching for the threshold of P that best explained the observed spatial distribution of DENV counts per state. We set 100 thresholds from 0.5 to 2.5 and measured the amount of time (T) the index P remained above each threshold in each state in 2015. Measures of T per state were used to calculate Pearson’s correlation with DENV counts per state (Figure 2D). We found that small thresholds (P>1) had non-significant and low correlations between yearly cases and mean yearly suitability, while intermediate-to-high thresholds (P>1.35) had significant and high correlations. The minimum correlation was found just above P=1 (dashed vertical lines in Figure 2D), when states are seen to spend equal amount of time T above that threshold in a year (seen by measures of T per state coloured according to DENV counts as in map’s legend, Figure 2D). These results show the amount of time a region spends with suitability above a typical threshold of 1 is the least optimal to predict yearly epidemic potential and that in Myanmar’s epidemiological context higher thresholds are more informative.

Discussion

In this study, we were able to demonstrate that our measure for mosquito-borne viral suitability is informative in the context of Myanmar, despite the lack of ento-epidemiological datasets with high spatio-temporal resolution. By estimating suitability through climate variables and known ento-epidemiological parameters, we were able to project mosquito-born virus (MBV) suitability at the district level, a resolution for which epidemiological data and mosquito abundance measures are not generally available. Here, we discuss the national and subnational public health and control implications for MBVs in the context of our projections.

At both the national and subnational levels, the wet season (Jun – Oct) was estimated to have the highest potential for MBVs transmission, in accordance with reported epidemiological time series. Since this was observed across Myanmar, it suggests that the epidemic potential of each district peaks during this period independently of their spatial location. In contrast, the hot and dry season (Mar – May) presented the lowest potential, also consistently across Myanmar, in accordance with what are known to be less favourable climate conditions for the vector. We therefore argue that in Myanmar, adequate surveillance and health care delivery resources should be fully operational by the end of the hot and dry season (May), in anticipation for the increase in MBVs case counts that is likely to occur in the following months. A similar argument can be used for vector control strategies in Myanmar, which should have maximal impact before the onset of the wet season, when mosquitoes encounter less favourable climate conditions and have smaller population sizes.

We also identified important spatial variations in MBV suitability across Myanmar. For instance, the highest potential was found primarily in the southern rural districts bordering with Thailand, and to a lesser degree across the middle of the country where the 3 major urban centres are located. Importantly, this estimated potential was highly correlated with DENV cases counts at the state level, although the lack of epidemiological data at such spatial resolution precluded us from verifying if our estimations at the district level fitted local counts. It may be tempting to speculate that the higher number of DENV counts in the south is a consequence of inflow of cases from Thailand (a highly endemic country). Critically, however, the estimated suitability in the south of Myanmar, in particular during the wet season, suggests for the first time that the observed higher incidence is driven by a local, higher intrinsic potential for MBVs transmission. Although management of case importation should be part of any national plan, it is clear that reducing suitability in the southern region of the country will be critical to control MBVs, for which vector control strategies should be effectve.

Epidemiological models are useful tools to gauge the burden of a disease of interest, assess transmission potential, mosquito suitability, prompt surveillance efforts, inform better public health policies and highlight areas for pressing research. Such approaches are even more critical in epidemiological settings characterized by the absence of sustained surveillance or for pathogens which tend to have mild or asymptomatic pathology. The method here introduced requires solely climatic data and basic ento-epidemiological assumptions for which literature support is available. We foresee the usefulness and applicability of the index P for other regions of the world for which surveillance data is still missing, either due to lack of resources or absence of a pathogen of interest. In contrast, for regions rich in historical spatio-temporal and ento-epidemiological data, the index P may be a starting point for the development of an early warning system which would be based on real-time input of climatic variables.

Implications for Zika virus in Myanmar

Our understanding of Zika virus (ZIKV) epidemiology in Myanmar and other countries of South East Asia (SEA) is incomplete, although there is evidence of continued transmission in the region from serosurveys and occasional viral isolation in residents and travellers to the region 31,32,33,34. Such evidence also supports the notion that ZIKV transmission in SEA preceded that of the Americas (2014-2015, 35) even with an apparently low number of cases and no major epidemics reported. To date, only one imported ZIKV case has been notified by the Myanmar Ministry of Health and Sports (MOHS) 36 and the virus’ spatio-temporal potential for transmission in the country is largely unknown. Given that mainland countries of SEA share many of the climatic and eco-demographic factors that dictate positive suitability for Aedes mosquitoes, it is reasonable to assume that regions within Myanmar have the potential for epidemic or endemic transmission of ZIKV. Exploiting the fact that transmission seasons of various Aedes-born viruses (e.g. DENV, CHIKV, ZIKV) tend to be synced in time in other regions of the world 37⁠, we here discuss and speculate on the ZIKV public health implications of the index P’s spatio-temporal patterns found both at the national and subnational levels in Myanmar.

The higher suitability for MBVs in southern Myanmar suggests the south to be a viable route for importation of ZIKV from Thailand. It is known that southern international borders are home to sizeable mobile populations with limited access to healthcare 38. Introduction of ZIKV through such borders would therefore carry a significant public health burden but would also likely be difficult to detect with a passive surveillance system. Additionally, suitability and DENV counts suggest a path of high transmission potential in the middle of the country, in districts including the 3 major urban centres of Myanmar (Yangon, Naypyitaw and Mandalay). Due to the domestic nature of the mosquito species involved, urban centres are a hallmark for ZIKV transmission and establishment, with attack rates above 60% reported elsewhere 15,39,40 . For the city of Yangon, for example, a similar attack rate would result in +3 million cases, and would incur significant health and economic consequences. Public health prevention or mitigation of a starting epidemic therefore calls for active surveillance initiatives that move beyond formal international points of entry (i.e. airports and maritime ports) and urban centres. Detecting early epidemic transmission chains in time for mosquito-control interventions before the wet season may effectively hamper the full potential of ZIKV and prevent high attack rates.

Exposure to ZIKV infection during gestation is a major risk factor for development of a variety of neonate neurological complications including microcephaly (MC) 35,41,42. Recent studies have further suggested that the risk of MC is highest for exposure around week 17 of gestation, resulting in a lag of approximately 5 months between ZIKV and MC epidemic peaks 15,41,42,43. Based on the estimated time window of peak MBV suitability in Myanmar between June and October, we therefore predict that, in the event of a ZIKV outbreak, an epidemic of MC in the country would occur between November and March. This time window is therefore critical for active MC surveillance to be established in Myanmar. To date, there has been no report of significant increases in MC cases in Myanmar. Caution should be taken in assuming this as evidence for no ZIKV circulation, since in previous epidemics there has also been a lack of reported ZIKV-associated MC cases ⁠44 and it is possible that only one of two existing lineages of the virus is responsible or such clinical manifestations 35,41,42,45 .

Put together, the estimated spatio-temporal variations in MBV suitability found in this study suggest that in order to decrease mosquito populations before the onset of ZIKV epidemics or prevent potential ZIKV introduction events from the southern region, control initiatives should take place just before, and at the beginning of, the wet season. Special attention could also potentially be stratified across districts or states in the middle of the country, including the major urban centres, and in particular in the south, as these regions are likely to have higher transmission potential.

Limitations and future work

There are certain limitations to our approach. We note that the unavailability of high resolution climatic data meant that (1) it was impossible to estimate suitability along the border with China and Laos, two countries in which DENV transmission is reported to be endemic; and (2) that weather stations had to be used for vast geographical ranges, limiting our capacity to explore potentially relevant spatial heterogeneities within the larger districts. The climatic data used was also limited to 2 years, and although we show that the index P in that period explains much of DENV’s epidemiology in 1996-2001, it is uncertain to what degree our estimations could have been better with matching time periods. It should also be noted that we take care in not interpreting P>1 as a critical threshold for transmission potential. The real epidemic thresholds (R0>1, Re>1) are dependent on the total number of female mosquitoes per human (M=NV/NH) which is largely unknown in time and space. In this context, our sensitivity analysis in Figure 2D helps to elucidate this and can be of use for other regions for which climatic variables are available. Another climatic data source that could be investigated in future is satellite remote sensing, although we did not use this for the present study. We also discuss the implications for ZIKV transmission in Myanmar, although our results are based on DENV epidemiological data. Given that no seroprevalence or epidemiological data exists for ZIKV in Myanmar, our discussion points are intended to inform the community to the best of our knowledge, but should be taken as speculative until new data is made available and compared to our current projections. Our approach also does not include demographic factors, why may affect both the human susceptibility and the vector carrying capacity. Although our index P can explain much of the spatio-temporal patterns of Myanmar, it is possible that these factors explain some of the missed spatial patterns (e.g. local vector capacity could be higher in regions with more cases than predicted by P).

Ento-epidemiological count data

DENV case counts for Myanmar between 1996 and 2001 (as published by 12⁠) were published already aggregated at the level of the country and by month. Naing et al. reported that the original source of the case counts was the official annual reports of the Myanmar National Vector-Borne Disease Control programme (VBDC). Cases included total suspected reports of dengue fever (DF) and dengue haemorrhagic fever (DHF). The absolute counts, per month (Jan-Dec) were: 20, 35, 40, 43, 139, 333, 487, 255, 258, 261, 80, 82 (year 1996); 67, 68, 37, 55, 140, 724, 876, 877, 545, 430, 248, 150 (year 1997); 131, 144, 208, 271, 714, 2511, 2904, 2455, 1528, 1163, 747, 249 (year 1998); 45, 105, 164, 91, 445, 1309, 1533, 968, 533, 247, 92, 102 (year 1999); 32, 29, 26, 90, 120, 333, 164, 95, 59, 73, 71, 84 (year 2000); 16, 29, 576, 1190, 2137, 2868, 3082, 2346, 1476, 990, 354, 51 (year 2001).

DENV counts at the level of Myanmar’s states for the year 2015 were used, as reported by the Ministry of Health and Sports 30. Incidence per 100k individuals per state were: Sagaing 125, Ayeyarwady 105, Mandalay 106, Mon 259, Yangon 63, Bago 68, Tanintharyi 145, Naypyitaw 163, Shan 50, Kayin 108, Magway 35, Kachin 61, Rakhine 28, Kayah 102, Chin 18.

Data collected in a study of Aedesaegypti abundance in Yangon over the year of 2011 was also used 29. From this publication we used the number of major breeding containers found in Yangon per month (Jan-Dec): 110, 102, 58, 51, 109, 249, 257, 276, 258, 248, 252, 170.

Spatial and climatic data

The administrative distribution of Myanmar into districts was suitable for our analysis, since it was possible to classify them by predominant weather conditions, using the Köppen-Geiger classification 46⁠: equatorial monsoonal (Am), equatorial winter dry (Aw), warm temperate-winter dry-hot summer (Cwa), warm temperate-winter dry-warm summer (Cwb) and arid steppe-hot arid (BSh). We obtained climate data from the United States National Oceanic and Atmospheric Administration webpage 47, which had incomplete observations that we then complemented with information from the Department of Meteorology and Hydrology, Yangon, for the period 2015-2016. Time and resource constrains for this process of data collection allowed for retrieving data from 14 weather stations, which were representative of the following districts: Pathein station, for the districts of Pathein, Pyapon, Maubin, Myaungmya and Labutta; Hpa An station, for Hpa An, Myawaddy, Kawkareik, Mawlamyine and Thaton; Sittwe station, for Sittwe, Marauk-U and Maungdaw; Dawei station, for Dawei, Myeik, and Kawthoung; Yangon Airport station, for North, South, East and West Yangon; Bago station for Bago, Hpapun and Hinthada; Nay Pyi Taw Airport station for North and South Nay Pyi Taw, Yamethin and Magway; Loikaw station for Loikaw, Bawlake and Langkho; Katha station for Katha, Bhamo and Mohnyin; Hkamti station for Hkamti district only; Taunggyi station for Taunggyi and Loilen; Mandalay Airport station for Mandalay, Kyaukse, Miyngyan, Nyaung-U and Meiktila; an average of the weather conditions in the Am region, for the districts of Kyaukpyu and Thandwe); and an average of the weather conditions in the Aw region, for the districts of Minbu, Pakokku, Gangaw, Pyinoolwin, Sagaing, Shwebo, Monywa, Kale, Yinmabin and Kyaukme.

To include a district in the present analysis, we used the criteria that its main population settlements were below 1500 meters above sea level, since the entomological modelling system we employed does not account for the effect of elevation on vector ecology and higher altitudes are less favourable to mosquito survival, plus either of the following: having access to climate variables from its weather station; or that its central point was within 100 Km of a station from which climate information was available; or being situated within a weather region were climate could be extrapolated from other districts’ stations. The latter was done since an analysis of variance showed no difference in mean temperature across weather stations in the Am (F=0.391, p-value=0.53; Dawei, Hpa An, Yangon, Pathein and Sittwe, and Bago stations) and Aw regions (F=2.793, p-value=0.09; Taungoo, Loikaw and Nay Pyi Taw stations). Extrapolation was not done for districts within the Cwa region, as there was a statistically significant difference in weather observations from individual stations (F=12.03, p-value>0.05; Hkamti, Katha nad Taunggyi stations). Lastly, Mandalay was a single station within the BSh region and Hakha station from the Cwb region was removed from analysis, due to elevation criteria.

The three weather seasons defined in this study for the context of Myanmar were: cool dry season, from November to February, hot dry season, from March to May, and wet (monsoon) season, from June to October. National means (standard deviations) of yearly and cool, hot and wet season were, correspondingly: temperature in degrees Celsius 27.1 (2.8), 26.7 (1.5), 28.8 (1.3), and 27.7 (1.4); percent humidity 77.9 (11.1), 79.9 (5.3), 64.5 (5.8), 85.3 (6.7); and inches of rainfall 0.17 (0.43), 0.03 (0.13), 0.01 (0.05) and 0.47 (0.62).

Parameters specific to Myanmar

As detailed in the main text, unknown parameters α, ρ and η were obtained for each weather station using a parameter sweep with heuristics of adult mosquito life-span of ~9 days and extrinsic incubation period of ~ 5 days (over the period 2015-2016). See Methods for details. The obtained values for α, ρ and η per weather station were (in order): Pathein 1.414, 0.78, 2.241; Hpa An 1.414, 0.78, 2.241; Sittwe 1.552, 0.45, 2.517; Dawei 1.414, 0.56, 2.517; Yangon 1.552, 0.78, 2.241; Am 1.414, 0.78, 2.379; Bago 1.414, 0.67, 2.379; Nay Pyi Taw 1.552, 0.45, 2.517; Loikaw 1.966, 0.45, 2.379; Aw 1.552, 0.78, 2.241; Katha 1.689, 0.56, 2.379; Hkamti 1.828, 1.00, 1.828; Taunggyi 1.828, 0.78, 1.828; Mandalay 1.552, 0.23, 2.655.

The districts found to have yearly mean index P>1 were: Magway, Nay Pyi Taw north, Nay Pyi Taw south, Yamethin, Bhamo, Katha, Mohnyin, Dawei, Kawthoung, Myeik, Maungdaw, Mrauk-U, Sittwe, Bago, Hinthada, Hpapun, Hkamti, Hpa-An, Kawkareik, Mawlamyine, Myawaddy, Thaton (with no particular order).

Data Availability

The ento-epidemiological count data used is of public domain, previously published in other studies, and is here made available (counts per date) in a dedicated subsection of the manuscript. The climatic data was obtained from the United States National Oceanic and Atmospheric Administration (USOAA) webpage with permission granted by the administration. The climatic data is owned by the USOAA. Access to the data can be granted by USOAA upon request (https://www.research.noaa.gov/Contact).

Funding

PNPG received travel and accommodation expenses from the Department of Global Health and Oriel College, University of Oxford. The Medical Action Myanmar (MAM) provided funding for purchasing the weather data from the Department of Meteorology and Hydrology. JL received funding from the European Research Council under the European Union’s Seventh Framework Programme (FP7/2007-2013)/ERC grant agreement no. 268904 – DIVERSITY. UO received an EMBO postdoctoral fellowship. The Myanmar Oxford Clinical Research Unit is part of the MORU Tropical Health Network, funded by the Wellcome Trust. RJM receives funding from Asian Development Bank and the Bill and Melinda Gates Foundation. Mahidol-Oxford Tropical Medicine Research Unit is funded by the Wellcome Trust of Great Britain. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing Interest Statement

Authors declare no competing interests.

Corresponding Author

José Lourenço (jose.lourenco@zoo.ox.ac.uk)

]]>
http://currents.plos.org/outbreaks/article/measuring-mosquito-borne-viral-suitability-in-myanmar-and-implications-for-local-zika-virus-transmission/feed/ 0
Recent Trends in Unpasteurized Fluid Milk Outbreaks, Legalization, and Consumption in the United States http://currents.plos.org/outbreaks/article/recent-trends-in-unpasteurized-fluid-milk-outbreaks-legalization-and-consumption/ http://currents.plos.org/outbreaks/article/recent-trends-in-unpasteurized-fluid-milk-outbreaks-legalization-and-consumption/#respond Thu, 13 Sep 2018 14:10:25 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=76143 Introduction: Determining the potential risk of foodborne illness has become critical for informing policy decisions, due to the increasing availability and popularity of unpasteurized (raw) milk.

Methods: Trends in foodborne illnesses reported to the Centers for Disease Control in the United States from 2005 to 2016 were analyzed, with comparison to state legal status and to consumption, as estimated by licensing records.

Results: The rate of unpasteurized milk-associated outbreaks has been declining since 2010, despite increasing legal distribution. Controlling for growth in population and consumption, the outbreak rate has effectively decreased by 74% since 2005.

Discussion: Studies of the role of on-farm food safety programs to promote the further reduction of unpasteurized milk outbreaks should be initiated, to investigate the efficacy of such risk management tools.

]]>

Introduction

Current information regarding the risks and benefits of unpasteurized (raw) dairy products is important for decisions regarding food safety policy, and is especially relevant given increasing demand for locally sourced, unprocessed foods, as well as in light of accumulating evidence for health benefits of consuming unpasteurized milk. While nutrient loss on pasteurization of milk is slight 1, cross-sectional studies from several countries have now reproducibly demonstrated significant beneficial immunological effects of unpasteurized milk consumption, primarily protection against childhood asthma, atopy and respiratory illnesses 2, 3, 4, 5, 6, 7, 8 . Several of these studies, including the PARSIFAL study of 15,000 children in five countries 9, 10, 11, and the GABRIEL study of 10,000 children in three countries 12, 13, 14, 15, 16 , were designed to identify determinants of the “farm effect” by which children exposed to farming environments have lower incidence of infections and allergic disorders. Further evidence for the preventive effects of raw milk exposure, both in utero and during infancy and childhood, was found from prospective birth cohort studies, notably the six country PASTURE birth cohort of over 900 children monitored until age 6. This study confirmed the strong protective effect of early raw milk consumption against asthma and atopic diseases, and provided insight into mechanisms of immunomodulation 17, 18, 19, 20, 21, 22, 23, 24, 25, 26. Recently, the protective effect of childhood raw milk consumption against adult asthma and atopy, and an association with higher adult lung function, has been demonstrated in the prospective Agricultural Lung Health study of over 3,000 farmers and their spouses in the United States 27, 28 . Studies on human cell lines and mouse models are being used to determine the relevant milk components and to further elucidate cellular and genetic mechanisms of action 29, 30, 31, 32 .

This new body of evidence suggests that, given the potential for significant public health benefits which could be gained from a reduction in immunological disorders, a re-evaluation of the risk/benefit profile of unpasteurized milk is in order. A growing number of American states have already legalized unpasteurized milk via farm-gate sales, retail sales, herdsharing, or licensed pet food sales, yet legislators must weigh growing consumer demand against the risk of outbreaks of illnesses due to milk-borne pathogens. Two previous studies 33, 34 analyzed state-specific outbreak and legalization data from the United States from 1993 to 2006 and from 2007 to 2012, respectively. Comparing reported outbreak rates in states which provide consumers with legal access to unpasteurized fluid milk vs. states without legal access, both studies concluded that further legalization of unpasteurized milk would likely result in increased numbers of outbreaks. As data on reported outbreaks through 2016 is now available from the Centers for Disease Control and Prevention (CDC), this paper continues the analysis.

Methods

Outbreak data collection and classification

A foodborne disease outbreak is defined as an incident in which two or more people experience a similar illness resulting from the ingestion of a common food 35. Outbreaks in the United States are reported by states to the National Outbreak Reporting System (NORS) of the Centers for Disease Control and Prevention (CDC) 36. Applications were submitted to request NORS outbreak data from 2005 to 2016, related to dairy foods and total outbreaks. The start date of 2005 was chosen as reporting might have been less complete during the first years of the eFORS system, which replaced paper forms in 1998. The field “EstimatedPrimary” provided values for number of illnesses associated with each outbreak. The field and value of “IFSACLevel2=Dairy” was used to extract a set of dairy-related outbreaks; note that this excludes outbreaks involving multiple food types (e.g. ground beef + fluid milk).

Dairy-related outbreak records were reviewed and classified as pasteurized or unpasteurized, processed or unprocessed, and whether a dairy species other than cow was involved. Additional information to clarify contaminated ingredients and pasteurization status was obtained as necessary from the CDC 37, 38, 39, state health departments 40, 41,42, 43, 44, Morbidity and Mortality Weekly Report 45, 46, 47, and a third-party website 48, 49, 50, 51, 52, 53, 54, 55.

Calculations related to outbreaks and legalization in this paper focus specifically on outbreaks involving unpasteurized fluid milk, excluding those related to processed dairy products (e.g. cheese). For purposes of this analysis, “processing” is defined as including methods which result in a change in consistency of the final product, including inoculation, incubation, condensation, and dehydration; but excluding pasteurization, homogenization, or the addition of flavourings. Unpasteurized milk and cheese as consumables are regulated differently under state and federal law. For example, unpasteurized cheese is legal for sale and interstate distribution if aged as specified in federal regulations 56, while laws governing the sale of unpasteurized fluid milk vary by state but interstate distribution is illegal 57 .

Data quality issues

NORS data for 2009 was excluded from outbreak trend analyses due to data quality limitations specific to that year. The CDC’s 2012 report National Outbreak Reporting System: An Evaluation of Foodborne Disease Outbreak Surveillance and Technical Requirements for Reporting notes that there was a decrease of almost half of the previous five-year average for all outbreaks, which may have been caused by: “1) technical issues associated with the introduction of the new system, 2) reassignment of types of outbreaks previously reported as foodborne to another mode of transmission, 3) staffing and budgetary issues … or 4) other, unidentified reasons” 58.

A second data quality issue involves errors, inconsistencies and omissions in NORS data. In record CDCID 15533 for example, the value for the field “ExposureState” is recorded as “Pennsylvania” whereas the correct value is “Multistate.” Of 27 records related to queso fresco cheese, six (CDCIDs 2275, 3822, 12355, 14760, 259206, and 268763) were recorded under “IFSACLevel3” as being “Fluid milk” and the other 21 as “Solid/semi-solid dairy products.” Other records contained blank fields or were ambiguous, e.g. in record CDCID 15143 the field indicating pasteurization status (IFSACLevel4) was blank and the implicated food is described simply as “milk.” Additional information was sought when necessary, as described above, and all corrections were confirmed with CDC staff and are detailed in Supplementary Table 1.

State population scaling

Annual population estimates from 2005 to 2016 were obtained from the U.S. Census Bureau for each of the 50 states plus the District of Columbia59, 60. DC has been included as a separate jurisdiction in calculations related to outbreaks and legal status for three reasons: it has a law prohibiting raw milk distribution 61; the CDC lists outbreaks for “Washington DC” as a separate “state”; and it has an estimated population (July 1st 2017 = 693,972) comparable to states such as Alaska (739,795), Vermont (623,657), and Wyoming (579,315).

Population-scaled outbreaks were graphed as outbreaks per million persons for each year, with polynomial regression line fitted in Excel.

Legal availability of unpasteurized milk

Information about state laws governing unpasteurized milk distribution was compiled from National Association of State Departments of Agriculture (NASDA) unpasteurized milk surveys for 2004, 2008, and 2011 62, 63, 64, state government websites 65, 66, 67, 68, 69, 70,71, 72, 73, 74, 75, 76, 77, 78, 79, and third-party sources 33, 80, 81,82, 83,84, 85, 86, 87, 88, 89, 90, 91, 92. A start date of 2004 was chosen as this was the date of the earliest comprehensive NASDA survey.

Five distinct conditions of legal status and distribution were identified and coded: Retail (R; legal off-farm retail and/or farm market sales), Farm-gate (F; legal farm-gate sales but no off-farm sales), Herdshare (H; herdshares permitted by law, written policy, or court order, but sales remain illegal), Pet food (P; farms holding a “pet food” or “commercial feed” permit may sell the product), and Illegal (I; both herdshares and sales are illegal). This system allows for a more precise analysis than the two-group (legal vs. illegal) classification system used by Langer et al. 33 and is less complicated than the seven-group system used by Mungai et al. 34.

Herdsharing is differentiated from selling as it involves legal co-ownership of a herd of dairy livestock by a group of consumers, and states such as Alaska and Colorado (category H) have legalized herdshares while maintaining a ban on sales 66, 70 . The 2004 NASDA survey lists herdsharing as being illegal in Alaska; however, this is incorrect as it was legalized in 1998 via a regulatory exemption. In addition to allowing sales, jurisdictions classified in this study as R, F, or P may also have some number of active herdshares distributing raw milk, either licensed and regulated by state law (e.g. Washington State) or unregulated and operating in a legal “grey area” (e.g. California).

It should be noted that the legal status of unpasteurized milk in a jurisdiction is not an exact indicator of accessibility. As an example, some private buying clubs distribute raw milk across state lines, including to jurisdictions such as Virginia, Delaware, and the District of Columbia, where sales are illegal 90.

Though not noted in the two fore-mentioned studies, the legal sale of unpasteurized milk as pet food or commercial feed is included as a separate category. This classification is noted because designation as pet food does not necessarily mean that the primary consumers are animals, as suggested by this Indiana Board of Animal Health statement:

Indiana has a pet food license issued through the State Chemist office at Purdue University. Many raw milk farms have increased sales of their pet food in the form of milk, butter, yogurt, and cream. 92

In 2008 in North Carolina, a proposed rule requiring dye to be added to pet milk resulted in a successful lobbying campaign by advocates to have the state legislature reverse the rule 93, 94. In 2016, the Florida State Department of Agriculture had 84 “registered feed distributors” selling unpasteurized fluid milk to the general public 95. This evidence from three states appears to indicate that classifying unpasteurized milk as “illegal” in states issuing licences for “pet milk” sales may not accurately reflect human consumption patterns within those states.

States were classified where possible according to the actual distribution situation in a state. Accordingly, Nevada has been classified as illegal (I) even though sales are technically legal in that state, as licensing requirements enforce a de facto prohibition. A county milk commission which includes a physician and a veterinarian must first be established, committee regulations must be approved by the state, and milk can only be distributed within that county. Only one county milk commission exists, and as of May 2018 no farm had received approval 96. Virginia was classified as Herdshare (H) rather than Illegal (I) as the state permits them on a case-by-case basis 86, with 67 listed on one consumer website 90. Indiana was classified as Herdshare although it also has legal provision for pet food (P), as this appears to be the more common distribution route: the same website lists 30 herdshare farms compared to 7 farms licensed to sell unpasteurized milk as pet food 90. Kentucky and Rhode Island both permit unpasteurized goat milk purchase with a doctor’s prescription, but there is no evidence of this being common practice, so both are classified as Illegal for years in which this was the only legal means of access. For simplicity, the year in which a jurisdiction changed status was assigned the new status.

Access to unpasteurized milk is defined here as “legal” for any jurisdiction categorized as retail, farm-gate, herdshare, or pet food (R, F, H or P).

Foodborne outbreaks may involve multiple states. For between-state comparisons in calculations involving outbreak rates and legalization, the specific states involved in multi-state outbreaks were noted as separate “state-specific outbreak incidents.”

Unpasteurized milk production licenses

Numbers of active licensees and permit holders were requested from all state agencies which issue licences and permits to farms to allow them to produce and distribute unpasteurized fluid milk. Where data was not available from state governments, secondary sources such as conference proceedings and media reports were used 63, 97, 98, 99, 100, 101,102, 103, 104, 105, 106, 107,108, 109, 110, 111, 112,113, 114, 115, 116. Missing data were estimated by interpolating from confirmed numbers.

Results

Total reported outbreaks

Over the twelve year period from January 1 2005 to December 31 2016, there were 10,965 reported foodborne disease outbreaks, resulting in 208,734 illnesses, 10,585 hospitalizations, and 233 deaths. Outbreak information for different food categories is displayed in Table 1. Food vehicles were only identified in 5,236 (48%) of all outbreaks (FB_FoodMain: FoodVehicleUndetermined = “FALSE”), with the foods implicated in the remaining 5,729 outbreaks (52%) undetermined. A caveat should therefore be made that this or any other comparative analysis which uses NORS outbreak data should be viewed in light of the fact that implicated foods remain undetermined in a large proportion of reported outbreaks.

Of outbreaks with identified food vehicles, 232 (4.4%) involved dairy foods (Supplementary Table 1), resulting in 9.2% of all illnesses.

Pasteurized dairy products were responsible for 32 dairy-related outbreaks (13% of all dairy), 2,225 illnesses (45%), 120 hospitalizations (26%), and 17 deaths (74%). Of the 232 dairy-related outbreaks, the pasteurization status of 17 could not be determined. Unpasteurized fluid milk was associated with 152 outbreaks (66% of all dairy), resulting in 1,735 illnesses (35%), 169 hospitalizations (38%), and two deaths (9%). Four of the 152 unpasteurized dairy outbreaks involved both fluid milk and cheese products made from the same milk. Five of the 152 outbreaks involved only goat milk and two involved both cow and goat milk.

Table 1 also displays the average number of deaths per thousand illnesses. Unpasteurized fluid milk shows a similar figure (1.2) to total foodborne outbreaks (1.1), pasteurized dairy is higher at 7.6, and dairy products which are both pasteurized and processed are significantly higher, at 40 deaths per thousand illnesses; this issue will be explored in a follow-up study.

Table 1: Reported Foodborne Disease Outbreaks: Total Outbreaks and Dairy Categories from 2005 to 2016. “Unpasteurized fluid milk” outbreaks include four outbreaks which also involved cheese made from this milk. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention.

Table 1: Reported Foodborne Disease Outbreaks: Total Outbreaks and Dairy Categories from 2005 to 2016. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention.

Twelve-year review of outbreak trends

Excluding data for 2009 (5 outbreaks), annual reported outbreaks related to unpasteurized fluid milk started at a low of 10 in both 2005 and 2006 and rose to peak at 18 in both 2010 and 2011. After this peak, outbreaks then saw a general decrease: 14 in 2012, 16 in 2013 and 2014, 11 in 2015, and 13 in 2016. This results in an annual average of 14 outbreaks for the most recent 5 year span, from 2012 to 2016 inclusive.

To analyze outbreak trends, it is necessary to control for changes in population size. When scaled using U.S. Census Bureau population estimates, outbreaks associated with unpasteurized fluid milk increased from 0.034 outbreaks per million persons in 2005 to 0.058 per million in 2010, then decreased to 0.040 per million by 2016. This equates to a 19% percent increase from 2005 to 2016, and a 30% decrease from 2010 to 2016 (Figure 1). The 2009 value (0.016) is less than half of the average of the preceding three and following three years, illustrating the noted data integrity issue for that year.

Figure 1 new

Figure 1: Outbreaks associated with unpasteurized fluid milk per million persons; plotted in Excel showing polynomial regression trend line. Outbreaks reported for 2009 provided to illustrate data quality issues specific to that year but excluded from trendline and further calculations. Data from the U.S. Census Bureau population estimates and the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention.

Outbreak rates and legalization of unpasteurized milk

Table 2 shows the five categories of legal availability of unpasteurized milk and the total proportion of the U.S. population in 2016 living in jurisdictions classified by each code. Complete information for each of 51 jurisdictions (50 states plus the District of Columbia) for 2004 to 2016 is displayed in Supplementary Table 2 using these codes.

Table 2: Legal availability of unpasteurized milk to consumers and percentage of the U.S. population living in states within that category (2016). Data from U.S. Census Bureau population estimates, National Association of State Departments of Agriculture unpasteurized milk surveys, state governments, and third-party websites.

Table 2: Legal availability of unpasteurized milk to consumers and percentage of the U.S. population living in states within that category (2016). Data from U.S. Census Bureau population estimates, National Association of State Departments of Agriculture unpasteurized milk surveys, state governments, and third-party websites.

Figure 2 displays a 14 year timeline from 2004 to 2017 of the total number of jurisdictions in each category as defined in Table 2. The total number of jurisdictions providing consumers with legal access to this product (R, F, H & P) increased from 32 in 2004 to 43 in 2017 (+34%); conversely, the number of jurisdictions with no legal distribution (I) fell from 19 to 8 (-58%). Jurisdictions permitting herdshares while banning sales (H) increased from 4 to 11 (+175%), off-farm sales (R) increased from 10 to 12 (+20%), while states permitting farm-gate sales (F) varied between 13 and 15. States where unpasteurized milk is mainly available as pet milk (P) increased in 2016 from 4 to 5 when Maryland began granting permits to farms 81. No state passed laws to restrict or remove accessibility during the study period.

Figure 2: National trends in the legalization of unpasteurized milk in the U.S. (2004-2017), applying the legalization categories defined in Table 2. Legal status data from state governments, National Association of State Departments of Agriculture (NASDA) unpasteurized milk surveys, and third-party websites.

Figure 2: National trends in the legalization of unpasteurized milk in the U.S. (2004-2017), applying the legalization categories defined in Table 2. Legal status data from state governments, National Association of State Departments of Agriculture (NASDA) unpasteurized milk surveys, and third-party websites.

In order to analyze the relationship between outbreaks and legalization within and between jurisdictions, the specific states involved in six multi-state outbreaks (CDCIDs 11065, 15533, 257838, 261453, 262220, and 268599) were identified (Supplementary Table 3). As an example, outbreak CDCID 262220 is associated with illnesses in three states: Illinois, Indiana, and Michigan. For purposes of examining outbreak rates vs. legalization within these states, this outbreak was recorded as involving three “state-specific outbreak incidents.” Excluding 2009 data results in the 156 state-specific unpasteurized fluid milk incidents used in further analysis.

Outbreaks per million persons from Figure 1 are plotted in Figure 3 against the number of states which provided consumers with legal access via sales or herdsharing during the twelve year period from 2005 to 2016. With a Pearson’s correlation coefficient of 0.26 (95% confidence interval: -0.40 to 0.74), analysis of the CDC outbreak data for 2005 to 2016 does not support the suggestion that increased legal access to unpasteurized milk leads to higher outbreak rates.

Figure 3: States permitting legal access to unpasteurized fluid milk compared to outbreak rates in the United States, 2005-2016. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Figure 3: States permitting legal access to unpasteurized fluid milk compared to outbreak rates in the United States, 2005-2016. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

To further examine the hypothesis that legalization leads to higher outbreak rates, one can examine whether a change in legal status within a specific jurisdiction affects the outbreak rate within that jurisdiction. The legal status of unpasteurized milk changed in eleven states during the 12 year period from 2005 to 2016. Six states where raw milk distribution was illegal legalized herdsharing (Kentucky, Michigan, North Dakota, Ohio, Tennessee, and West Virginia), two states with farm-gate sales legalized retail sales (South Carolina and Utah), one illegal state legalized pet food sales (Maryland), one state with legal herdsharing then legalized farm-gate sales (Illinois), and one illegal state first legalized herdsharing and then legalized farm-gate sales three years later (Wyoming).

Five of these states (Kentucky, Michigan, North Dakota, Tennessee, and Wyoming) legalized unpasteurized fluid milk distribution during the middle part of the study period, such that at least four years of outbreak data are available prior to and following legalization (excluding 2009 due to data quality issues), classifying the year in which legalization occurred as “After”. As the subset of data is small, outbreak-incident rates for these five states have been averaged over the four-year before and after periods, shown in Table 3. There was no change in absolute number of outbreaks (6 outbreaks in each 4 year period) and a slight reduction in relative outbreak rates due to population growth during this time.

Table 3: Outbreak incidence rates per million persons before and after legalization of unpasteurized fluid milk in five states for which there is at least four years of outbreak data prior to and after legalization. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Table 3: Outbreak incidence rates per million persons before and after legalization of unpasteurized fluid milk in five states for which there is at least four years of outbreak data prior to and after legalization. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Figure 4 shows the average annual outbreak incidence rate per million persons in the U.S. by legal status of unpasteurized milk. Significant variability is seen both over time and between categories of legalization, in part representing fluctuations due to the small absolute number of outbreaks per year within each category. While the Retail category shows the highest overall level of outbreaks, there has been a clear downward trend since 2012. Similarly, the outbreak rate for Farm-gate states peaked in 2011 and has remained low since 2012. The highest variability is seen in the Herdshare and Illegal categories, potentially due to the lack of regulatory oversight in these states.

Figure 4: Outbreak incidents per million persons by legal status of unpasteurized milk, 2005 to 2016 excluding 2009. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Figure 4: Outbreak incidence rate per million persons by legal status of unpasteurized milk, 2005 to 2016 excluding 2009. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

This summary of outbreak rates by legal status of the state (Figure 4) masks variation between states within each category of legalization. Table 4 illustrates this variation by stratifying states according to average annual outbreak rate over the 12 year period. Within each of the four levels of frequency of outbreaks, there is wide variation on legal availability. Twelve states did not report any outbreaks from 2005 to 2016, six of which provide legal access to unpasteurized milk. While variability is necessarily expected due to the small absolute numbers of outbreaks reported per state, as above, these variations demonstrate that the legal status of unpasteurized milk is clearly not the only determinant of outbreak rate within a state.

Variation also occurs over time within states. As an example, Vermont had the highest average annual outbreak rate at 0.436 outbreak incidents per million persons; however, Vermont had no reported outbreaks from 2011 to 2016. Similarly, Washington State (0.083) had no reported outbreaks from 2012 to 2016.

Table 4: States stratified by average annual outbreak incident rates for 2005-2016 with 2009 excluded. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Table 4: States stratified by average annual outbreak incident rates for 2005-2016 with 2009 excluded. Outbreak data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention. Population data from the U.S. Census Bureau.

Outbreak rates vs. consumption trends

It is necessary to examine the relationship between outbreak rates and consumption trends in order to determine whether the decline in outbreak rates after 2010 may be due simply to a decline in consumption.

The CDC FoodNet “Atlas of Exposures” 2006/2007 survey indicated that 3.0% of the American population had consumed unpasteurized milk in the previous 7 days 117. However, this survey has not yet been repeated to allow a comparison, nor is there a national reporting system for unpasteurized milk production. It is therefore necessary to use other quantitative measures to estimate trends in consumer demand. One such proxy is the number of state-issued unpasteurized milk farm licences and permits.

Annual totals for the number of licences and permits issued were requested from the fifteen state government agencies which license or issue permits to farms to produce and distribute unpasteurized fluid milk. Adequate data was received from nine of these agencies, representing a diverse geographic area and 40.7% of the total 2016 U.S. population (Table 5). Missing licensing data were estimated by interpolating from confirmed numbers, as indicated by square brackets.

Table 5: Number of licences and permits issued to unpasteurized milk farms in nine U.S. states, 2005-2016. Estimates indicated by square brackets. Legal status indicated by colour code as in Table 2. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention, and state government licensing information.

Table 5: Number of licences and permits issued to unpasteurized milk farms in nine U.S. states, 2005-2016. Estimates indicated by square brackets. Legal status indicated by colour code as in Table 2. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention, and state government licensing information.

Compiling licence and permit numbers for these nine states shows an increase over twelve years from 76 to 347 licensed unpasteurized milk dairies (+357%). Assuming that licence and permit numbers are a reasonable proxy for consumption, and factoring in U.S. population growth (9.3% over the twelve year period), the ratio between outbreak rate and consumption rate shows a pronounced decline, with the 2016 outbreak-to-consumption ratio only 26% that of 2005 (Figure 5).

Figure 5: Estimated trend in outbreak rates controlling for population growth and estimated consumption rates, 2005-2016, scaled to show values relative to 2005. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention, U.S. Census Bureau population estimates, and state government licensing information.

Figure 5: Estimated trend in outbreak rate controlling for population growth and estimated consumption rate, 2005-2016, scaled to show values relative to 2005. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention, U.S. Census Bureau population estimates, and state government licensing information.

This calculation represents only an approximate indicator, as only nine states are represented, not all farms producing unpasteurized milk in each state might be licensed, not all licensing agencies maintain an ongoing record of the total number of current active licensees, and changes in farm-specific production levels are not reflected. As an example of how licence numbers may underestimate consumption rates, Missouri reports only one active unpasteurized milk licence 101 while two consumer websites advertise 16 118and 88 current sources of unpasteurized milk within that state, respectively 118, 90. As an example of farm-specific change, the Organic Pastures Dairy Company increased production by approximately 60% between 2006 and 2016 (email from Mark McAfee, mark@organicpastures.com, October 5, 2016):

In 2006, we were in 400 stores and 8 farmers’ markets and produced about 680,000 gallons of raw milk per year, versus 2016 with 700 stores, 22 farmers’ markets and 1.09 million gallons .

These examples suggest that our calculation may well underestimate consumption growth, such that the actual decline is likely to be more pronounced than what is shown in Figure 5. The Pearson’s correlation coefficient between number of states with legal access and consumption-scaled outbreak rates of -0.83 (95% confidence interval: -0.96 to -0.47) indicates that, contrary to predictions of previous studies, increased access to and production of unpasteurized milk in the United States over this twelve year period has shown a strong negative correlation with foodborne outbreak rates.

Discussion

Comparison with related studies

By analyzing data from 2007 to 2012, a previous study concluded that “Legalization of the sale of unpasteurized milk in additional states would probably lead to more outbreaks and illnesses” 34. The current study extends that analysis by expanding to cover a twelve year period from 2005 to 2016, controlling for population growth, examining outbreak trends in jurisdictions in which legal status or availability of the product has changed, and estimating changes in consumption during the study period.

In contrast with other studies 33, 119, this paper does not include outbreaks caused by unpasteurized cheese. Not only is unpasteurized cheese production and distribution regulated under Federal law and is thus unrelated to state-specific laws regulating unpasteurized fluid milk distribution, but processing itself introduces other factors which could affect relative risk, an issue which will be explored in a separate analysis.

Also in contrast to other studies 34, 119, this study excluded 2009 outbreak data from calculations due to data quality issues specific to that year 58. Suggestive of this issue, the figure of nine dairy-related outbreaks in 2009 is less than half of the average annual number (μ=20.3, σ =1.86) for the remaining 11 years in the 12 year period under observation. Using 2009 data in longitudinal analyses of trends could therefore suggest a larger increase relative to that year than what actually may have occurred.

Outbreaks and Legalization

Given the four observed trends of a reversal in the number of reported outbreaks (Figures 1 and 3), increasing legalization (Figures 2 and 3), no increase in outbreak rates in five states which legalized raw milk (Table 3), and increasing consumption (Table 5), evidence was not found that supports the position that the legalization of unpasteurized milk within a jurisdiction will cause an increase in outbreaks. Indeed, examining data up to and including 2016 shows that increased legal access after 2010 has been concurrent with generally declining outbreak rates, irrespective of change in consumption. As Figure 4 and Table 4 both illustrate, outbreaks can still occur in states where distribution is illegal, so legal prohibition itself is not a guarantee of consumer safety.

Using the proxy measure of number of licensed unpasteurized milk dairies, consumption was estimated to have increased by 357%. Although this is only suggestive of a general upward trend, a similar increase was recently seen in the United Kingdom, where raw milk consumption in England, Wales, and Northern Ireland increased from 3% in 2012 to 10% in 2018 120.

A decline in outbreaks caused by a particular food vehicle could be due to reduced consumption of that food or to changes in methods of production, processing, and handling. As an example, the 1997 introduction of the HACCP (hazard analysis and critical control points) food safety system in meat processing plants throughout the U.S. was responsible for a 42% decline in E. coli-related illnesses over the subsequent seven years 121. Similarly, the decline in unpasteurized milk related outbreaks could be related to changes in product handling or to the implementation of bacterial testing standards and inspection programs. Further studies should explore the efficacy of factors such as education and regulatory programs in preventing outbreaks, and in particular, examine what may have enabled some states (e.g. Vermont and Washington) to reduce outbreak rates to zero while others (e.g. Utah) have seen no decrease (see Supplementary Table 4).

Regarding education, a challenge for dairies has been a lack of on-farm food safety programs. This changed in 2010 when the Farm-to-Consumer Legal Defense Fund made training materials available and the Raw Milk Institute began developing a HACCP-based on-farm food safety program 122, 123, 124. The decline in frequency of outbreaks coincides with the introduction of these targeted education programs. The tentative conclusion can be drawn that, similar to what was seen in the meat processing industry, the implementation of on-farm food safety systems for unpasteurized milk production may be related to the observed reduction in outbreak rates.

Data from Pennsylvania supports this connection. As shown in Figure 6, outbreaks occurred each year from 2006 to 2014, then no outbreaks were reported for 2015 or 2016, then one outbreak occurred in 2017 125. An interesting correlation is that in November 2014, Pennsylvania State University’s College of Agricultural Sciences hosted a workshop on unpasteurized milk safety in collaboration with the Raw Milk Institute 126, 127. In addition, in 2014 the proprietor of the largest Pennsylvania unpasteurized milk farm was trained and listed with the Raw Milk Institute 128. This farm had been responsible for outbreaks in 2012 and 2013 (CDCIDs 15533 and 15482) including one associated with 148 out of 258 (57%) unpasteurized milk related illnesses reported nationwide in 2012 129, 130, 131. No outbreaks have been associated with this farm’s products since training and listing. While it is impossible to show causation, this correlation of fewer outbreaks with the implementation of a HACCP-based on-farm food safety training program has implications for further studies of the role which education and extension programs may play in the safe production and handling of unpasteurized milk.

Figure 6: Annual number of outbreaks and illnesses related to unpasteurized fluid milk reported in Pennsylvania, 2006-2017. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention and Pennsylvania State Bureau of Epidemiology.

Figure 6: Annual number of outbreaks and illnesses related to unpasteurized fluid milk reported in Pennsylvania, 2005-2017. Data from the Foodborne Disease Outbreak Surveillance System, Centers for Disease Control and Prevention and Pennsylvania State Bureau of Epidemiology.

It should be noted that this study does not examine whether or not states which legalized raw milk also implemented mandatory licensing, bacterial testing standards, on-farm food safety plans, or farm inspections by state agriculture or health departments as part of the legalization process. Regulatory programs such as these may well mitigate outbreak rates, and further studies should explore this question. For example, Vermont had three outbreaks between 2008 and 2010, but has had no incidents since. This change coincides with an adjustment period after the introduction in 2009 of a stronger regulatory system 132.

The federal ban on interstate trade 57 in unpasteurized fluid milk has not eliminated either multi-state outbreaks or the emergence of unregulated interstate buying clubs. As 43 states representing 92.6% of the U.S. population currently (March 2018) permit legal access to unpasteurized milk, and legalization does not correlate with increased outbreak rates, expanded legalization along with the development of a federal regulatory framework could be considered, similar to national systems already in effect in other nations such as England, France, Germany, and New Zealand 133, 134, 135, 136.

Study Limitations

This study is subject to the limitations of the underlying data, as CDC does not have a record of all outbreaks which have taken place in the United States. Not all outbreaks are identified, investigated, or reported, and the source of foodborne illnesses are often not identified. Even when a single food is identified, the point of contamination is not always known or reported. As documented by the CDC, “The quality of NORS data is dependent on outbreak investigations conducted by state and local health departments, the usability of the reporting system, and the resources available to each state and local health department to investigate and report outbreaks.” 58 Nevertheless, the CDC database contains a substantial body of evidence with which to estimate true outbreak rates, and as such, is widely used for public health policy decision-making. In addition, state bacterial standards were not examined in order to determine if these have an effect on outbreak rates, and it is recommended that this be the subject of a future study.

A more significant limitation relates to the difficulty in estimating the actual extent of unpasteurized milk production or consumption. It is apparent from licensing statistics and consumer websites that unpasteurized milk is currently being produced on a larger scale than in past decades; however, with the lack of standardized reporting, and with many farms operating outside of any regulatory structure, any estimate of unpasteurized milk production necessarily incurs a large margin of error. The CDC FoodNet “Atlas of Exposures” survey provided the most comprehensive estimate of unpasteurized milk consumption published to date, but without data beyond 2006/7, there is no record of recent trends.

Further Studies

As the variability of state outbreak rates is not simply dependent on legal status, other factors beyond the scope of this study might be affecting the safety of unpasteurized milk production. Topics for further investigation could include: state-specific regulatory structures including inspections and licensing; farm-specific factors such as product handling and pathogen testing; consumer education regarding food safety and the appropriate transport, storage and handling of unpasteurized dairy products; and comparison of milk produced by dedicated raw milk farms with milk from conventional dairy farms doing “incidental sales” of unpasteurized milk to the public. Together, such studies would provide valuable tools to aid in assessing the impact of policy decisions and state or federal unpasteurized milk regulatory structures, thus facilitating evidence-based decision making in public health.

Conclusions

From a public health perspective, the lack of consistency and comprehensiveness in measuring production or consumption of unpasteurized milk is problematic; however, this analysis provides a current best estimate of the scale of disease outbreaks due to unpasteurized milk. The potential for foodborne illness continues to be a small but real risk from consuming unpasteurized fluid milk, but analysis of data over a twelve year period demonstrates that increased access to this product within the United States has not led to increased outbreak rates. On the contrary, total reported unpasteurized milk-associated outbreaks have declined since 2011, despite increased production, and outbreak rates proportional to estimated consumption rates have declined by 74% over the twelve year period.

The evidence that legalization of unpasteurized milk has correlated with decreased outbreak rates has potential implications for public policy decisions. Recent introduction of on-farm food safety training programs for unpasteurized milk producers may be a factor in the recent decline in outbreak rates. Further studies of the efficacy of such “best-practices” training will be necessary in determining the utility and efficacy of these risk-management options, and could enable the transition from prohibition-based to harm reduction-based regulatory structures. This in turn will enable the further development safe and minimally processed dairy products, to take advantage of the enormous public health benefits that would result from a significantly lower incidence of infections and allergic disorders provided by consumption of fresh, unprocessed milk.

Data Availability

This study is based on publicly available data from the U. S. Centers for Disease Control and Prevention’s National Outbreak Reporting System (NORS) and the United States Census Bureau. The legal status of unpasteurized milk was determined from National Association of State Departments of Agriculture (NASDA) unpasteurized milk surveys, state governments, and third-party websites. Licensing data were obtained from state governments. Details of all dairy-associated outbreaks, as well as regulatory status and outbreak rate for each jurisdiction and year in the study period are available in the Supplementary Materials.

Supplementary Information

All supplementary material is available at http://figshare.com/s/866c3d82f50105ff5dab

Corresponding Author

Joanna Whitehead is the corresponding author and be contact at jowhite@uvic.ca.

]]>
http://currents.plos.org/outbreaks/article/recent-trends-in-unpasteurized-fluid-milk-outbreaks-legalization-and-consumption/feed/ 0
Information Circulation in times of Ebola: Twitter and the Sexual Transmission of Ebola by Survivors http://currents.plos.org/outbreaks/article/information-circulation-in-times-of-ebola-twitter-and-the-sexual-transmission-of-ebola-by-survivors/ Tue, 28 Aug 2018 15:30:11 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=82825 Introduction: The 2013-2015 outbreak of Ebola was by far the largest to date, affecting Guinea, Liberia, Sierra Leone, and secondarily, Nigeria, Senegal and the United States. Such an event raises questions about the circulation of health information across social networks. This article presents an analysis of tweets concerning a specific theme: the sexual transmission of the virus by survivors, at a time when there was a great uncertainty about the duration and even the possibility of such transmission.

Methods: This article combines quantitative and qualitative analysis. From a sample of 50,000 tweets containing the words “Ebola” in French and English, posted between March 15 and November 8, 2014, we created a graphic representation of the number of tweets over time, and identified two peaks: the first between July 27 and August 16, 2014 (633 tweets) and the second between September 28 and November 8, 2014 (2,577 tweets). This sample was divided into two parts, and every accessible publication was analyzed and coded according to the authors’ objectives, feelings expressed and/or publication type.

Results: While the results confirm the significant role played by mainstream media in disseminating information, media did not create the debate around the sexual transmission of Ebola and Twitter does not fully reflect mainstream media contents. Social media rather work like a “filter”: in the case of Ebola, Twitter preceded and amplified the debate with focusing more than the mainstream media on the sexual transmission, as expressed in jokes, questions and criticism.

Discussion: Online debates can of course feed on journalistic or official information, but they also show great autonomy, tinged with emotions or criticisms. Although numerous studies have shown how this can lead to rumors and disinformation, our research suggests that this relative autonomy makes it possible for Twitter users to bring into the public sphere some types of information that have not been widely addressed. Our results encourage further research to understand how this “filter” works during health crises, with the potential to help public health authorities to adjust official communications accordingly. Without a doubt, the health authorities would be well advised to put in place a special watch on the comments circulating on social media (in addition to that used by the health monitoring agencies).

]]>

Introduction

Social media today are key sources of data that can be used for understanding health crises. They are, for users, a space and a hub for circulating information 1,2,3 and, for researchers, a tool for understanding how health crises are experienced and perceived 4,5,6,7,8,9. The fact that users may share information or promote rumors and misinformation 10,11,12,13,14 offers unprecedented opportunities for studies to address the links between the public’s posting of information (original thoughts, sharing and questioning of material from both mainstream media and official sources) and levels of fear and panic 15,16,17,18,19,20. As a free application with contents restricted to 140 characters, the networking and microblogging platform Twitter is one of the most popular social media. During the most recent Ebola outbreak, approximately 100,000 tweets containing the word “Ebola” were posted each day before September 30, 2014 – the date that the first Ebola diagnosis on American soil was reported. Once news of this diagnosis had “broken”, this number multiplied by eleven, first reaching around 870,000 tweets by the end of the day, and 1.6 million tweets the day after, on October 1st 2014, across Europe, North America, Africa, Asia and Australia 21. Similar spikes occurred following official announcements of Ebola cases in the United States 8, despite the actual small number of cases.

Most information posted on social media during a health crisis has been shown to emanate from the accounts of news organizations and health authorities, but also from humorous accounts and even celebrity ones 11. Recent studies have called on health authorities to monitor Twitter, to identify the most active and influential accounts, and to understand or even to adopt social media platforms and rhetoric 11,20,22. This is all the more important that it is now evident that Twitter has become as an important public space for criticism of health authorities: Roberts et al. 20 showed that official health information had some prominence in mainstream media reports but less in social media. In other words, public health officials have failed to impose their narrative during the crises.

However, if it is clear health authorities need to learn from social media users, we are yet to understand precisely how these users contributed to shape the debates that emerged online. By analyzing in detail the sentiments, questions and critiques that users expressed in tweets, on the one hand, and their reading and circulation of mainstream media articles and official announcements on the other, this study provides new insights on the social conversation that arose during the most recent Ebola crises. A more precise comprehension of the diverse and sometimes contradictory perceptions of these crises is crucial to a better understanding of the circulation of information during a humanitarian crisis.

Our study combines qualitative and quantitative analysis and focuses on a specific topic discussed on Twitter: the sexual transmission of Ebola by survivors. This topic was controversial at the time of the outbreak, as the nature of such transmission was not widely understood. An overview of studies on the subject shows few mentions 21,23,24,25 until the first transmission was confirmed on March 20, 2015, after which numerous articles were published 26,27,28,29,30,31,32. It is now known that Ebola virus genetic material can linger more than 500 days in semen 33. Subtil et al. 34 estimate that the probability of Ebola RNA is up to 31.6% at 3 months and falls to 0.7% at 18 months. Same decreasing trend was noticed by Sow et al. 35. Moreover, the variability of Ebola RNA presence in semen has been highlighted by Keita et al. with patients testing positive, then negative, and then positive again 36. Although the World Health Organization (WHO) published a Twitter post on March 23rd 2014 stating that sexual transmission of Ebola was possible via “infected semen” for “up to seven weeks after clinical recovery”, this news, although an official pronouncement, had little impact on Twitter. Only one user repeated its content the same day. Our data shows that it was only at the end of July that the issue began to gain importance online. We examined two significant peaks identified in a sample of 50,000 tweets that were posted between March 15 and November 8, 2014. The peaks had significantly different contexts. The first one started on July 27 and ended on August 16, 2014 (633 tweets). It relates to the fast spread of the virus to countries like Nigeria, and its consequences such as the ban that South Africa and Kenya (and indeed Liberia) imposed on travelers from West Africa, as well as the preparations of the United States and Europe to deal with a potential outbreak. The second peak started on September 28 and ended on November 8, 2014 (2,577 tweets). It followed the dying of the first Ebola patient diagnosed on American soil, Thomas Eric Duncan. At the same time as these peaks were occurring, the total number of cases was declining and the WHO was hoping to announce that Nigeria and Senegal were free of Ebola after 42 days with no new cases.

Our results show that while mainstream media continued to strongly influence the information in circulation, online debates became autonomous on the matter of the sexual transmission by survivors. This led to the expression of criticisms of official institutions and about the definition of Ebola as an STD, with strong references to AIDS. At a first level of analysis, the results confirm the significant role played by mainstream media in diffusing information. But at a second level, and contrary to what is usually understood in the literature, our data show that Twitter does not fully reflect other media content. Rather, it works more like a filter that highlights issues that have been published by mainstream media. By analyzing the circulation of information on Twitter and the responses of its users (be these responses criticisms, jokes or expressions of doubt), this article not only complements existing scientific literature but also provides insight into the social debate that occurred around the Ebola epidemic. It analyzes which information was in circulation and how it was summarized, reformulated, criticized or simply rejected online. In addition, this study examines how Twitter users can, during a debate on a subject as sensitive as the sexual transmission of the virus, act as complementary to official sources by autonomously selecting and highlighting an issue that was treated as a side-note by the media. More broadly, this article attempts to show how an analysis of posts sent through Twitter during outbreaks might be used in order to adjust official communications.

Materials and Methods

Data collection

Web content mining methods were applied to the tweets with steps similar to those demonstrated by Yoon and Bakken 37: selection of keywords; importation of data; preparation of data; analysis of data; content analysis (topics and sentiments) and finally, data interpretation. Inspired by previous studies 37,38,39,40, we first used French and English keywords to identify relevant tweets. These two languages are widely used by people from the different countries affected by Ebola (Guinea, Liberia, Sierra Leone, and secondarily, Nigeria, Senegal and the United States). Semiocast, our partner company for social media analysis, provided the first sample of 50,000 tweets containing the words “Ebola”, posted between March 15 and November 8, 2014, allowing us to preprocess the data 40. For ethical and technical reasons, it was impossible for us to locate Twitter users and categorize them according to this parameter. All tweets were extracted without any personal data to comply with ethical rules for studies on public health issues. We used a topic model to identify the latent issue of sexual transmission by survivors 41: two researchers analyzed this sample and created a list of 17 keywords, keeping the best possible equivalence between French and English, in order to extract all tweets, with the exception of retweets without modification (Table 1).

OBK-17-0034 Table 1

Table 1: French and English keywords.

These two researchers then manually sifted through this sample and removed irrelevant tweets. In cases of uncertainty as to classification, tweets were discussed collectively. When necessary, we used the Twitter search application to understand the discussion context. The final corpus numbered almost 6,000 tweets, making it impossible to carry out a full analysis. We created a graphic representation of the number of tweets over time, and identified two peaks: the first between July 27 and August 16, 2014 (633 tweets) and the second, between September 28 and November 8, 2014 (2,577 tweets). An information watch on all health institutions’ Twitter accounts (ONG, CDC, WHO, Public Health Agency of Canada) was also set up during the research to monitor their announcements and reactions from the Twittersphere.

Classification

This sample was divided into two parts, and each researcher coded half the data. When the classification of a tweet was unclear, the tweet was discussed collectively. The coding categories were not defined beforehand but elaborated according to the characteristics of the data, in order to take the polysemy into account.

The tweets were clustered according to the authors’ objectives (see, for example, reference 42). Five categories were created:

  • Asking for information;
  • Challenging and rectifying information;
  • Providing information;
  • Giving an opinion;
  • Other.

We also indicated tweets expressing feelings, such as anger, laugher or fear. We took into account only tweets with hashtags (Such as #weregonnadie, #funfact, #fail, #whathasobamadonetous, etc.), emoticons or clear indications about a specific feeling (Such as “we all dead”, “It’s only getting worse”, “LOL” “hahahahaa”, “LMAO”, “That’s a lie!”, etc.).

As in previous studies (such as the study carried out by Hughes and Palen 43), we took any URL links into account to analyze the “networked interplay” (20) between information from mainstream media and tweets. Every accessible publication was analyzed and coded according to publication type. Five categories of publication were identified:

  • Newspaper articles from mainstream media with an editorial byline (The New York Times, lemonde.fr, The Washington Post, etc.);
  • Publications from public organizations (ONG, CDC, the WHO, Public Health Agency of Canada, etc.);
  • Publications from blogs, social networks, discussion forums and video sharing websites (Youtube, Twitter, Facebook, MSN, etc.);
  • Scientific articles;
  • Other (specialized websites about health, medical care, travel, or online encyclopedia).

Results

The relative influence of mainstream media and official organizations on Twitter

98.7% of the tweets collected are written in English. But French tweets profiles are similar in terms of organization and general content, and there appeared to be no significant difference between English and French tweets.

Relatively few tweets express emotion. From the full corpus, 7.5% of the tweets clearly express anxiety or fear, 4.4% express laughter and 1.4% express irritation and anger. Some users show interest in their followers and 4.2% of the tweets provide advice:

  • Ebola can be transmitted through sperm too now!?? Ladies, DO NOT town! (August 13, 2014)
  • Ebola can last in semen for up to 3 months. For you singles don’t trust nobodyyyy (September 30, 2014)
  • Ebola can live in semen for, 3 months wrap it up kids (November 1st, 2014)

A minority of the textual messages in tweets request information (4.1%), express an opinion (4.1%), or either challenge or correct information (0.6%). As is consistent with previous studies, the majority (more than 90%) only provide information.

A large proportion of the tweets (48.6%) contain at least one URL (1,572 out of 3,234 tweets). Among them, 466 tweets (29.6%) are drafted similarly and contain only one headline and one URL (and sometimes the publication’s name). This format is automatically provided by numerous publication websites:

  • TIL that when males are cured of ebola virus, they can still transmit the virus in their semen for up to 2 mon… http://tinyurl.com/lzt6gks (August 4, 2014)
  • Ebola Might Be Sexually Transmitted? http://wp.me/p2iq4c-9wL (October 3, 2014)
  • Ebola outbreak: Survivors told to use condoms to prevent virus spreading http://ht.ly/2OKIWM (October 8, 2014)

It is very close to what has been described, in a study of the practices of Twitter users sharing information, as “slavish tweets.” This is a reference to the term “slavish copying” in copyright infringement cases, i.e. the Twitter users share information, staying totally neutral, without any comments 44.

Interestingly, the websites of health organizations are under-represented as sources of content. During the first peak, the WHO website accounts for only 9 links (5% of the URLs). During the second peak, in the ranking by website URL shared in the tweets, the WHO comes eighth (4% of all URLs) and CDC comes tenth (2.6% of all URLs). URLs link to a variety of media but the relative frequency with which these media are cited is very different: 1.4% (20 out of 1,417 valid URLs) are specialized sites and scientific articles, 5.8% (83 out of 1,419) are news aggregators, 8.1% (115 out of 1,419) are websites belonging to public organizations, 18.6 % (264 out of 1,419) are blogs, social networks or forums, and 65.9% (935 out of 1,419) are mainstream media.

The results reveal the strong influence of mainstream media. But mainstream media did not create the debate. Twitter interest in sexual transmission by survivors exceeded mainstream media coverage: the first publication dedicated to this topic, and shared on Twitter, was the article entitled “Ebola survivors can infect others with their sperm” published by the South African website Health 24 on August 13, 2014. Half of the 83 publications from blogs and mainstream media that specifically mention it, and were shared on Twitter, were published between October 5 and October 11, 2014. But the first peak on Twitter began at the end of July, showing that Twitter users did not wait until sexual transmission hit the headlines. They took this specific piece of information from a variety of sources, whether dedicated to a general presentation of the virus (Such as “Ebola virus disease” by Wikipedia), to the spread of the disease (Such as “Un modèle mathématique de la progression d’Ebola créé à l’EPFZ” published on October 8, 2014 by Romandie.com) or to survivors in general (Such as “Surviving Ebola: For those who live through it, what lies ahead?” published on July 30, 2014 by CBS). Some of these sources, such as “What Is Ebola? Six Things You Need to Know” (NBC July 29, 2014) or “How not to catch Ebola” (BBC October 7, 2014), contain only one or two sentences about sexual transmission. In other words, Twitter users deliberately selected this piece of information, mainly from mainstream media publications. The microblogging platform seems to work both as a filter of information but also as an alternative pool of information that users re-prioritize on Twitter.

Single tweets about sexual transmission by survivors, March–November 2014

Fig. 1: Single tweets about sexual transmission by survivors, March–November 2014.

Gaps between journalistic informations and conversations on Twitter

Of the publications shared on Twitter via an URL, 17.3% are not dedicated to the sexual transmission, showing that tweet authors valued this piece of information more highly than the journalists. A question remains: what is the purpose of the publications shared on Twitter?

In general, the publications shared on Twitter via a URL express concern about the spread of the virus, and its arrival in Northern countries, as shown in the following tweet. A typical example states that “The threat of the disease traveling to a distant country is very real. A single infected passenger on an aircraft could spread the disease to another continent” (The Anti Media “Preventative Measures for Ebola in Case of an Outbreak” July 29, 2014). Some reflected a degree of anxiety. The authors emphasize the risk, the severity of the virus and the ability of Ebola to travel by air. This anxiety is not specific to forums, social networks or blogs, as the following example shows: “It’s frightening, mysterious and yes, it could come here. […] No matter where you live, instant jet travel has made any infection capable of spreading worldwide, and that includes Ebola” says a journalist from NBC (“What Is Ebola? Six Things You Need to Know” July 29, 2014).

During the first peak (between July and August) numerous authors insist on the importance of avoiding bodily fluids, which are the main means of transmission. Avoiding sexual relationships with survivors appears to be the last recommendation. For example, in the article entitled “Ebola 101: The Facts Behind a Frightening Virus” (NPR, July 10, 2014), the sexual transmission of the virus is discussed in brackets: “How does it spread? Through close contact with infected blood, saliva, urine, stool and vomit. […] Treatment: […] Patients are declared Ebola-free if they don’t show any symptoms for several days and if repeated tests for the virus in their bloodstreams come back negative. (The virus can still linger in semen for months […])

During the second peak (between September and November), the number of cases was decreasing and the post-outbreak situation was discussed. Sexual transmission of the virus seemed to be somewhat more prominent in the mass media:

  • “Sex could keep the Ebola epidemic alive even after the World Health Organization (WHO) declares an area free of the disease” (Reuters “Male Ebola survivors told: Use a condom”, October 7, 2014).
  • “Many of the survivors right now have detectable amounts of Ebola virus DNA in their semen or vaginal secretions—and many of them, presumably, are having sex” (The Daily Beast “Ebola Might Be Sexually Transmitted” September 4, 2014).

But the articles’ authors tend to mitigate their remarks regarding survivors. Many of them condemn the stigmatization of survivors:

  • “The doctor has beaten the odds and survived Ebola, but he still has one more problem: the stigma carried by the deadly disease” (AP “Survivors of Ebola face second ‘disease’: stigma” April 27, 2014).
  • “Ebola survivors are likely to be shunned and isolated by their communities” (NPR “Ebola 101: The Facts Behind A Frightening Virus” July 10, 2014)
  • “These fears are almost entirely misplaced” (Mother Jones “How Long Does the Ebola Virus Survive in Semen?” October 8, 2014)

Only one article, entitled “Ebola survivor infects wife to death” (The Pluto Daily October 14, 2014), casts responsibility on a survivor, alleging that he killed his wife by imposing unprotected sexual intercourse upon her. Moreover, the author does not generalize this case, and he concludes by mentioning the important role of survivors, recalling that they “should be role models in society, as they should be the ones telling people about the realities of the disease rather than infecting others”.

Moreover, the articles’ authors present this information as a scientific uncertainty. 87.3% of the publications that focus on the sexual transmission of Ebola use the WHO as the main source of information, quoting a sentence from the WHO website, published on October 6. The WHO stressed the hypothetical nature of this information, stating that “more surveillance data and research are needed on the risks of sexual transmission, and particularly on the prevalence of viable and transmissible virus in semen over time.” Similarly, the media take into account this scientific uncertainty:

  • “there is still the potential for them to spread the disease” (Health 24 “Ebola survivors can infect others with their sperm” August 13, 2014)
  • “it’s unclear just how great a risk the semen of surviving men poses in the weeks following their illness” (Mother Jones “How Long Does the Ebola Virus Survive in Semen? “ October 8, 2014)
  • “No one is certain that the viral DNA is actually living, transmissible virus” (The Daily Beast “Ebola Might Be Sexually Transmitted” September 4, 2014).

In contrast, tweets do not always show such a cautious approach. Particularly, if we remove neutral tweets from the corpus, the remaining tweets reveal criticisms and questions about the authorities’ crisis management performance. Contrary to what one might expect, in general, tweets do not endorse the idea of survivors as a threat. The article entitled “Ebola survivor infects wife to death” (The Pluto Daily October 14, 2014) had limited uptake: it appears in only 4 links out of 1,236 valid URLs. In this sample, Twitter users show little interest in the survivors themselves: they express little blame and no compassion toward them. Only two tweets are direct accusations against survivors:

  • @cnn African MEN are ebola death guns spewing the virus in their semen up to 6 months after infection. Small wonder it is spreading so fast (October 9, 2014)
  • All Ebola bomb survivors are at jack in the box (October 9, 2014).

Tweets focus instead on the veracity of official information on the sexual transmission of the virus. Almost all the questions (131 tweets) ask for confirmation or additional information. Moreover, 40 tweets (85% of the tweets expressing anger) voice the idea that the risk posed by survivors’ ability to transmit the disease has not been communicated at all, or not on a wide enough scale, or contradicts previous official statements stressing that only sick people can transmit the disease:

  • #ebola Why didn’t CDC reveal that semen of Ebola patients transmits virus up to “7 weeks after clinical recovery”? http://www.phac-aspc.gc.ca/lab-bio/res/psds-ftss/ebola-eng.php (October 1st, 2014)
  • @CDCgov Wrong! Men who survive #Ebola have it in their semen for 7 weeks, yet they aren’t symptomatic then. #fail (August 2, 2014)
  • Ebola can live 74 days in semen! The government is lying to you – you do not have to touch someone just be anywhere near where they’ve been (October 8, 2014).

A few salacious jokes reveal difficulty with understanding how that information has been obtained:

  • Whoever is tasting the semen for Ebola is doing the work of angels. Wait, that’s not how they test it? (October 2d, 2014)
  • You tested

More pertinently, 10 tweets focus on condom use, which is denounced as ineffectual in the light of AIDS history:

  • Survivors told to use condoms(why not refrain from sex for 3 mo instead,DONK!) http://www.enca.com/africa/ebola-survivors-told-use-condoms via @eNCAnews (October 8th, 2014)
  • Ebola survivors told to use condoms? Great, it was such a success with HIV/AIDS wasn’t it! http://www.independent.co.uk/news/world/africa/ebola-outbreak-survivors-told-to-use-condoms-to-prevent-virus-spreading-9782058.html (October 8, 2014)
  • “ebola survivors should abstain from sex or use condoms for three months” seems idiotic to me, how’d you like to have that condom break (October 12, 2014)

Of the tweets expressing criticism, 54.3% attack the WHO, the CDC and the mainstream media. A further 8.6% blame President Obama and his government, while 37.1% of tweets are generic accusations and do not single out anyone in particular. Thus, as was found in research on the H1N1 epidemic 33, the CDC, the WHO and the mainstream media are the scapegoats of choice, as they are supposedly responsible for providing reliable information. Paradoxically, tweets requesting information are not generally addressed to them: governmental and health institutions (such as the WHO or the CDC) are mentioned in 30 tweets (22.4%) and only 18 tweets (13.4%) are sent to a mainstream media. The recipients of the tweets requesting information are mostly individuals (56%), half of whom have identified themselves as journalists, safety specialists or physicians. This behavior may be interpreted as a sign of mistrust in official institutions (names from personal account have been deleted)

Understanding Ebola as an STD: the prevalence of AIDS

Another question is much more developed in the tweets than in the publications: is Ebola an STD? Only a few articles question the type of the disease once it was suggested the virus might be transmitted by sexual relation. One journalist writes that “it is also an STD of sorts” (Quartz “Here are the 35 countries one flight away from Ebola-affected countries” July 30, 2014). Similar mentions included “Ebola is never going away, WHO warns, Ebola IS a sexually transmitted disease” according to a headline in Catholic Online (August 10, 2014) and “Yes, the Ebola virus is potentially a sexually transmitted disease” confirms The Daily Beast (“Ebola Might Be Sexually Transmitted” September 4, 2014).

Only four publications make a connection between AIDS and the sexual transmission of Ebola. The articles make it clear that the two diseases are widely feared and in both cases, condoms should be used to reduce the risk of transmission (Daily Beast “Ebola Might Be Sexually Transmitted” September 4, 2014; Government slaves “Ebola Might Be Sexually Transmitted?” no publication date; Pissin’on the roses “Until Gaëtan Dugas or Other Flying Rats Catch Ebola, The North American Risk Remains Low” April 1st, 2014). However, their authors do not significantly develop the comparison between the two diseases.

In contrast, the connection between Ebola and AIDS is more prevalent in the tweets. In 99 tweets (6% of the non-neutral tweets) Ebola is described as an STD or a parallel is drawn between AIDS and Ebola. These tweets rarely use quotations from media: the ideas expressed seem to come from the users themselves. From a qualitative point of view, the parallels between the two diseases are much more developed in the tweets than in the online publications. Twitter users often noted that the sexual mode of transmission is similar in both cases and condoms are necessary to prevent infection. But more specifically, AIDS is used to evaluate the gravity of Ebola. AIDS is thus treated as a kind of barometer, whose gravity is presumed known by all, used to gauge the dangers of this new virus:

  • #Ebola, a sort of STD #virus it makes #AIDS look mild. Transmitted in semen as late as 61 days after patient recovery http://qz.com/242388/here-are-all-the-35-countries-one-flight-away-from-ebola-affected-countries/ (July 31st, 2014)
  • People who recover from ebola still carry the virus, in sperm etc. #worsethanaids? (August 7, 2014)
  • @Ask Ebola spread by semen? * This could be worse than AIDs. What color is the ribbon? Red? (October 13, 2014)

Although people make connections between AIDS and Ebola in general, they rarely discuss their own sexual activities and prefer to discuss general perspectives. Out of the whole corpus of tweets, only 151 tweets (5% of the single tweets) refer to daily life in the message or by using hashtags (#StayStrapped, #metalpantswithlockon, #Abstain, #WrapItUp, #SafeSex, etc.). Jokes reveal that sexual transmission is a difficult subject to speak about:

  • “Ebola is not airborne mom!” -me “Ebola stays in sperm for two months!” – mom #onlymymom #ohgosh (October 18, 2014)
  • This girl literally out of nowhere in class goes “did you know the Ebola virus can be spread through semen?” (August 6, 2014)
  • Kid in my class: “Ebola is transmitted by semen so better keep your mouth closed.” (October1st, 2014)

Moreover, joking about sexuality is used to transform a serious illness into a funny subject according to a cathartic logic, in order to reduce the anxiety associated with the Ebola name:

  • Ebola can live in vaginal fluid for 3 months, Michael Douglas’ head emerges from his wife’s puss. *I beat cancer, I’ll beat this.* #Ebola (October 8, 2014)
  • “Male survivors may be able to transmit the disease via semen for nearly two months.” that’s a seriously long orgasm, folks. i want #Ebola (October 17, 2016)
  • The 50 Best Sex Positions For An Ebola Outbreak: (50 pictures of people not having sex, because ebola can live in semen for 60 days) (August 4, 2016)

Discussion

The study of Tweets confirms the influence of mainstream media, which represent a large majority of the publications shared on Twitter. In our corpus, Twitter was rarely used for disseminating an alternative media discourse. Interestingly, however, the microblogging platform does not reflect mainstream media contents in a servile manner. Users focused on the sexual transmission of Ebola months before the media coverage of this piece of information: the first peak began at the end of July 2014 on Twitter. Users attached greater importance to the sexual transmission of Ebola than the journalists did. In this sense, Twitter seemed to work more like a filter than like a booster.

Because of this particularity, some of the issues raised by the sexual transmission of Ebola are much more developed in the tweets – in jokes, questions and criticism of communication by the authorities – than in the publications shared. Even if a majority of tweets are neutral and only provide information, the feelings or questions expressed in them could help to identify misunderstandings during health crises.

Limits

The analysis is limited somewhat by the fact that it is impossible (for reasons of scientific ethics) to identify the location and socio-economic profile of individual Twitter users. We only worked on the discourses revealed in the Tweets, and could not investigate social groups. Nor can we give a reliable explanation of why English tweets are more numerous than French tweets. We were unable to discover why interest in the sexual transmission of the virus by survivors began at the end of July, and why not before, with the first tweet of the WHO, or after, with the media coverage.

It is also worth mentioning that we started an analysis of the images contained in tweets, only to realize that very few of them actually referred to the sexual transmission or even to the disease. Ebola is rather used as a metaphor for everyday life. We also used emoticons and emojis to help classify tweets according to the emotions they convey, but further research on these discursive marks would help to further understand the rhetoric of users online.

Finally, information is not only shared on Twitter. Another study could highlight the differences between the use of Twitter and that of comments or internet publications (forums, online articles, other social networks such as Facebook, etc.).

Conclusion

This article increases our knowledge of the relationship between Twitter and traditional media in times of epidemics. In contrast to previous studies that underlined how users rely on Twitter for sharing information from mainstream media (such as 1,2,3), our study shows that tweets combine information dissemination with emotional stances and critical views, leading to a new or greatly increased debate about a subject that was treated as insignificant by other media. In that sense, Twitter acts mostly as a filter of information as well as a space where it is reconstructed. In the “Twittersphere,” the issue of sexual transmission of Ebola by survivors was mostly voiced by citizens and not by mainstream media. They did so by emphasizing the potential danger of the transmission and by defining Ebola as a STD.

While many Twitter users tweet and retweet in a neutral way, merely sharing URL links from media or official sources, there is a strong active minority that otherwise uses its public speaking power. By expressing their fears and doubts, and using the argumentative resources of irony or comparison (over time, or with events considered similar), Internet users raise questions and express objections that the health authorities did not consider necessary to discuss. Without a doubt, the health authorities would be well advised to put in place very quickly, on the occasion of every health crisis, a special watch on the comments circulating on social media (in addition to that used by the health monitoring agencies). It could prove a uniquely useful way to identify, even in the form of weak signals, criticisms or worries to which it is better to give answers before the rumors increase.

Online debates can of course feed on journalistic or official information, but they also show great autonomy, tinged with emotions or criticisms. Although numerous studies have shown how this can lead to rumors and disinformation, our research suggests that this relative autonomy makes it possible for Twitter users to bring into the public sphere some types of information that have not been widely addressed. Our results encourage further research to understand how this filter works during health crises, and suggests the potential of such research (particularly by shedding light on what interests or alarms citizens) to help public health authorities to adjust official communications accordingly.

Corresponding Author

Celine Morin: morin.celine@gmail.com

Data Availability

Data are freely available: https://doi.org/10.6084/m9.figshare.6026399.v1; https://doi.org/10.6084/m9.figshare.6026405.v1.

Competing Interests

The authors have declared that no competing interests exist.

]]>
Tracing Back the Source of an Outbreak of Salmonella Typhimurium; National Outbreak Linked to the Consumption of Raw and Undercooked Beef Products, the Netherlands, October to December 2015 http://currents.plos.org/outbreaks/article/obk-17-0057tracing-back-the-source-of-an-outbreak-of-salmonella-typhimurium-national-outbreak-linked-to-the-consumption-of-raw-and-undercooked-beef-products-the-netherlands-october-to-december-20/ Thu, 16 Aug 2018 11:45:08 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=80930

Introduction. On 23 October 2015, six related cases with gastroenteritis called the Netherlands Food and Consumer Product Safety Authority. They suspected filet américain, a raw beef spread, to be the source of infection. Leftovers and stool samples tested positive for Salmonella Typhimurium. Multiple locus variable-number of tandem repeat analysis (MLVA) revealed a MLVA pattern (02-23-08-08-212), which had not been detected in the Netherlands before. Concomitantly, an increase of this MLVA type was observed in the national Salmonella surveillance, amounting to 46 cases between 26 October and 9 December.

Methods. To investigate whether filet américain or an alternative (related) source could  be linked to surveillance-reported cases, cases (n=38) were invited to complete a questionnaire and upstream source tracing to map the food supply chain was initiated.

Results. Rapid interdisciplinary action resulted in identification of a contaminated 46-ton batch of beef distributed via a Dutch deboning plant as the likely source of infection. In total, 24/29 respondents (83%) could be linked to the incriminated batch of beef products (predominantly filet américain and minced beef).

Discussion. Repeated identification of raw meat products as a source of infection emphasizes the importance of awareness of the risk of infection when handling or consuming these products. Improved measures and procedures on product labelling, pre-treatment or product testing should be considered.

]]>

Introduction

Salmonellosis is the second most reported zoonosis in the EU after campylobacteriosis1. Approximately 85% of human salmonellosis cases result from consumption of contaminated food, such as undercooked eggs, raw meat products, or raw fruit and vegetables1,2.

In the Netherlands, salmonellosis is only notifiable in case of a cluster with two or more human cases probably linked to contaminated food or drinking water3. Annually, approximately 15-20 Salmonella outbreaks are detected in the Netherlands4. Since 1987, fifteen regional public health laboratories together form the Dutch laboratory surveillance network for gastroenteric pathogens, which covers approximately 64% of the Netherlands3,5. As part of the Salmonella surveillance, these laboratories sent their Salmonella isolates to the National Institute for Public Health and the Environment (Rijksinstituut voor Volksgezondheid en Milieu, RIVM), where the National Salmonella Centre performs serotyping on submitted Salmonella isolates of human, animal or environmental origin.

In 2003 and 2012, the two largest recent national Salmonella outbreaks (~540 and 1149 laboratory-confirmed human cases) could be linked to contaminated eggs (S. Enteritidis) – imported during a large avian influenza virus outbreak – and salmon (S. Thompson)4,6. In 2009, raw beef products were epidemiologically strongly associated with an outbreak of S. Typhimurium with a novel phage type (23 cases), but microbiological analysis of incriminated meat products could not confirm the association7.

On 9 November 2015, an outbreak of Salmonella Typhimurium with a unique multiple locus variable-number of tandem repeat analysis (MLVA) pattern (02-23-08-08-212) was identified.

An outbreak investigation was started with the following aims (i) to find out whether cases of the local cluster and cases identified through the national surveillance shared a common or related source of infection and (ii) to trace the origin of the Salmonella contamination to prevent further cases and new outbreaks.

Methods

Case definition

As Salmonella Typhimurium with MLVA pattern 02-23-08-08-212 had not been previously detected in the Netherlands, a case was defined as a person with laboratory-confirmed diagnosis for the outbreak Salmonella type. The sixth case within the local cluster was not laboratory-confirmed, but included as probable case due to the strong epidemiologic link with the other cases in the cluster.

Microbiological investigation

Strains isolated from human samples and from food products were detected and isolated by medical microbiological laboratories and the Netherlands Food and Consumer Product Safety Authority (Nederlandse Voedsel- en Warenautoriteit, NVWA), respectively. The method used for the food samples is based on the NEN-EN-ISO 6579/AI, which is an amendment to ISO 6579 published in 2007. Additionally to this method, the first enrichments (BPW) were screened for the presence of Salmonella based on the presence of the invA gene, followed by transferring positive samples to the second enrichment (MSRV). Different from the method in the amendment, selective plating media BGA and MLCB, instead of XLD, were used for the isolation of Salmonella. This method, including the screenings PCR, has been validated for horizontal use in food matrices.

At the RIVM, strains were serotyped based on O- and H-group antigens according to the World Health Organization (WHO) Collaborating Centre for Reference and Research on Salmonella standards8. S. Typhimurium were also typed by means of MLVA, which can discriminate between S. Typhimurium strains9.

DNA testing of meat

A Salmonella source attribution model, as described earlier10, pointed towards pigs/pork as the most likely source of S. Typhimurium with this MLVA type. To investigate whether filet américain might have been contaminated with pork, we tested the leftovers for the presence of pig DNA using real-time PCR with the Biorad CFX 96 system.

Epidemiological investigation

The surveillance-reported cases were invited to complete a questionnaire, either by phone or by mail conducted by the regional Public Health Services. As the index cluster provided a strong lead towards the most likely source of infection, we used a questionnaire tailored to consumption of various meat products [beef (raw ingredient of filet américain) and pork (indicated by source attribution model)] and venues where meat products were purchased (such as butchers or supermarkets). The questionnaire comprised 12 pre-listed supermarket chains. All questions focused on the 7 days prior to the onset of symptoms. Other items of the questionnaire included sections on demography, clinical symptoms, date of onset and duration of gastrointestinal illness, hospitalization and travel history. Data from questionnaires was entered into Microsoft Access (Microsoft Office Professional Plus 2010, Microsoft Corporation, Washington, USA) and analyzed using descriptive epidemiology. The analyses were conducted in SAS for Windows (SAS Institute Inc., Cary, NC, USA, version 9.3) and Microsoft Excel.

Ethical considerations

Medical Research Ethics Committee review and approval were not required as these kind of investigations are part of the routine public health response to an outbreak of salmonellosis. Cases were not asked for explicit informed consent, but indicated their willingness in participation in the interview. A case register including personal identifiers was used during the outbreak, which was needed for outbreak response and action. This register was only accessible by the outbreak management team who also performed all analyses.

Trace-back

During outbreaks, the NVWA is responsible for food trace back investigations according to Directive 2003/99/EC11. Supply chains of possible sources are examined based on available evidence from microbiological tests and/or epidemiological investigations. As decided in the EU Regulation (EC) 178/2002 (General Food Law), it is required that food products are traceable in two directions: one step forward (e.g. to the customer) and one step backward (e.g. to the supplier) in the food chain12. Immediately after the typing results revealed S. Typhimurium isolated from filet américain consumed by the local cluster, the NVWA launched an investigation to trace-back the origin of the contamination and surveyed whether contaminated meat products were still on the market. The investigation aimed to map the food supply chain from retail/catering premises to processors/slaughterhouse, to assess production hygiene, ‘Hazard Analysis and Critical Control Points’ (HACCP) and traceability.

Results

Start of the outbreak

On 23 October 2015 (see timeline in Figure 1), a citizen living in the Southern part of the Netherlands, called the NVWA to report that in total six persons fell ill with gastrointestinal illness after having consumed filet américain, a bread spread consisting of finely chopped raw beef mixed with a herb sauce. A stool sample from one case of the cluster tested positive for Salmonella spp. On 29 October, stool samples from four of the five remaing cases tested positive for S. Typhimurium. On the same day, the NVWA detected S. Typhimurium in the filet américain leftovers, informed the regional Public Health Service ‘Hart voor Brabant’ who subsequently reported this local cluster to the RIVM.

On 9 November, typing results showed that the Salmonella isolates from the local cluster and the food leftovers had a unique MLVA-pattern (02-23-08-08-212). Furthermore, 26 additional cases were identified in that week. To check whether other European countries experienced a concurrent identical outbreak or had had an outbreak that could be linked to this one, an urgent inquiry was placed on the Epidemic Intelligence Information System for Food- and Waterborne Diseases and Zoonoses (EPIS-FWD) operated by the European Centre for Disease Prevention and Control (ECDC). This yielded no clues for solving the outbreak.

Fig1_Timeline of the outbreak of Salmonella Typhimurium in the Netherlands

Fig. 1: Timeline of the outbreak of Salmonella Typhimurium in the Netherlands, October to December 2015.

NVWA = Dutch Food Safety Authority; RHS = Regional Public Health Service; RIVM = National Institute for Public Health and the Environment

Between 26 October and 9 December 2015, 45 outbreak cases were laboratory-confirmed (including five from the index cluster of six cases). For 34 cases, the date of onset of illness was known (Figure 2); the majority (n=33; 97%) developed symptoms between 14 October and 21 October, whereas one case fell ill on 27 October. One case (4%) reported having been abroad for one day. The cases identified through the national surveillance (n=40) were distributed across the country. The majority of the laboratory-confirmed cases were female (n=26; 58%) and ages ranged from 2 to 79 years (median 21). Of these, 15 cases were 9 years of age or younger (33%), six cases were aged 10-17 years (13%), 16 were aged 18-49 years (36%) and eight cases were 50 years or older (18%).

Fig4_

Fig. 2: Epidemiological curve showing 34 cases with Salmonella Typhimurium infection by date of onset of illness in the Netherlands during October 2015.

Six cases belong to the case cluster (5 thereof laboratory-confirmed), whereas the remaining depicted 28 cases – who completed a questionnaire – were identified through the national Salmonella surveillance system. For twelve other cases involved in this outbreak, date of onset of illness was not provided.

Information from questionnaire

The six cluster cases were not interviewed, as the only common source they have shared eating was filet américain and already a microbiological link with the filet américain was found for this cluster. All other cases typed until 23 November (n=38) were invited to fill in the questionnaire. Among those, 29 responded (response rate 76%). All 29 cases reported gastro-intestinal symptoms, eight (28%) were hospitalized. Diarrhoea lasted between 3 and 21 days (median: 9 days), based on information provided by 21 cases (72%).

Consumption of filet américain was most often reported (76%), followed by minced beef (55%). In total 26 cases (90%) ate one or both products in the week before illness. In a continuous survey ongoing since 200813, around 20% of the Dutch general population reports eating filet américain in the past week, around 55% minced beef and 62% one or both products. Twenty cases (69%) reported to have purchased any meat products as part of their grocery shopping at a supermarket, two cases (7%) bought meat exclusively at the butcher (7%) and seven (24%) cases bought meat at both venues. Information on beef product, purchase date and purchase site was sent to the NVWA for the trace-back investigations.

Trace-back investigations of food products

The local cluster cases reported having purchased filet américain at supermarket chain D. Supermarket chain D produced filet américain locally by mixing prepared ground beef and a ready-made herb sauce. Inspection of the involved local production site did not reveal any breach of hygiene standards. Trace-back of the contaminated batch led to meat producer 1 (P1; Figure 3). P1 reported delivering 70% of its total meat supply to supermarket chain D. Raw meat linked to the contaminated filet américain was distributed to 300 branches of this chain. All raw material for filet américain at supermarket chain D originated from P1. These tracing efforts could link another 15 cases to supermarket chain D.

Fig3

Fig. 3: Flow of trace-back operations of raw material contaminated with Salmonella Typhimurium, the Netherlands, October to December 2015

The raw material distributed by producer P1 was traced back to a Dutch deboning plant, which had processed ~58 tons of meat parts on 8 October 2015, originating from six slaughterhouses from four different European countries (Figure 3). In total, the deboning plant delivered ~46 tons of meat for human consumption to 55 meat-processing plants; of these, 32 were located in eight other EU Member States (~15 tons). Using information from forward and backward tracing, together with information from the questionnaires, we could establish links between the Dutch deboning plant and suppliers of three other supermarket chains, a diner and delicacy shop to which nine additional cases could be linked.

The NVWA – via juridical ordinance – commissioned the plant on 1 December 2015 to inform all customers that received parts of the contaminated batch. Dutch customers of the deboning plant were legally required to inform the NVWA how much of the contaminated meat they had processed and distributed. As the contaminated 46-ton batch of beef was mixed with meat from other parties at various customers, the contaminated batch amounted to 400 tons in total. Of those 400 tons, around 65 tons (16%) were classified as risk products defined as “ready to eat” products or products very likely to be consumed with a mild treatment, insufficient to eliminate or reduce the risk of infection by Salmonella to an acceptable level taking into account consumers’ consumption habits14. Most risk products had already been sold and/or the best before date had expired. About 1 ton was still available and withdrawn to prevent further cases.

Microbiological testing of food products

Besides the positive leftovers, samples of filet américain and separate samples of sauce and raw meat meant to be used for filet américain were taken at producer P1. All samples were from other batches than the incriminated batch, and tested negative for Salmonella spp. No samples were taken at the deboning plant as the particular batch was already distributed.

Although the source attribution model indicated that the outbreak MLVA type was more likely to be linked with pork/pigs (58%), than with beef/cattle (3%), no pig DNA was detected in the leftover filet américain.

Discussion

Here we describe the investigation of a national outbreak of S. Typhimurium with MLVA type 02-23-08-08-212 linked to raw and undercooked beef products. Based on microbiological and epidemiological evidence, beef sold as filet américain and minced beef was the source of infection. Timely involvement of the different authorities and exchange of information regarding epidemiological, laboratory and trace-back results led to the identification of a 46-ton batch of beef. The batch was distributed via a Dutch deboning plant. The plant received its meat from slaughterhouses outside the Netherlands, and thus outside Dutch authority, and with Salmonella not being a food safety criteria on farm level, further trace back was not performed. As parts of this batch could be traced forward to eight other EU countries, an alert in the Rapid Alert System for Food and Feed (RASFF) was placed. Risk products that were still on the market were withdrawn to protect consumers.

‘Exposure to raw beef’ and ‘consumption of undercooked meat’ are known risk factors for S. Typhimurium infection15. In the Netherlands, filet américain constituted the likely vehicle of infection in three previous S. Typhimurium outbreaks7,16. Likewise, filet américain was also associated with outbreaks caused by Shiga toxin-producing Escherichia coli O157 in the past17,18,19. A previous Dutch outbreak caused by S. Typhimurium was linked to an outbreak in Denmark, which preceded the outbreak in the Netherlands16. In Denmark, cases were infected through consumption of carpaccio. Trace-back led to a contaminated batch of beef imported from Italy20. As the Dutch outbreak occurred after the Danish outbreak, this example shows that rapid source tracing and timely retraction of contaminated products from the market is vital to prevent further cases.

Although trace-back revealed a batch of 46 tons as contaminated, only 46 cases were reported. The identified batch is the production of one day in the deboning plant. It is therefore likely that the batch will not be contaminated completely or evenly due to a point source contamination of one or more carcasses. Secondly, the number of cases will be an underestimation as most people with gastroenteritis do not visit a physician or are not tested. It was estimated earlier that around 1 in 20 Salmonella patients will be laboratory-confirmed25,26 leading to approximately 900 outbreak cases in the general population.

In outbreak situations where the source of infection has a short shelf life, microbiological detection of the pathogen in the food source is often not possible as the product is no longer on the market when trace-back efforts commence. In the current outbreak, the alert by the local cluster allowed timely microbiological testing of leftovers of filet américain, which expedited the process of identifying the underlying source. Parts of the contaminated batch of filet américain were still on the market (expiry date products producer P1: 17 October 2015) and could be recalled. Besides early identification of an outbreak, collaboration between regional and national authorities, and public health and food authorities is imperative to find the source. Regular contact and exchanging of gathered information proved to be crucial. Trace-back efforts could link 24 of 29 interviewed national surveillance-reported cases to the contaminated batch of beef. Although leftovers cannot be regarded as formal evidence for the source of infection, the microbiological results provided a strong clue for the trace-back, which was later confirmed by the positive results of a survey done independent from the outbreak investigations. An animal welfare organization performed microbiological tests in mini hamburgers included in a package containing a selection of different types of meat for raclette grilling. These mini hamburgers tested positive for S. Typhimurium with identical MLVA type. Raw material used for this product was subsequently linked to the 46-ton batch distributed via the deboning plant (Figure 2).

The moment the national outbreak was detected, the local cluster could be linked to filet américain. Furthermore, the MLVA pattern had not been seen earlier in the Netherlands. Therefore, we refrained from conducting a classical case-control study and decided to tailor the questionnaire to meat products for two purposes. First, to confirm beef products as source of the outbreak and secondly, to gather information on type of beef products, purchase dates and purchase sites to feed the trace-back. This contributed to a timelier start of the trace-back investigation and a likely quicker withdrawal of incriminated meat. To check whether filet américain or other beef products indeed were eaten more often than expected, we used the national controlsurvey13. Although less accurate than a case-control study as the reports are not from the same week in time, it answered the need of confirmation. The increase in cases observed in the regular surveillance would have similarly resulted in an outbreak investigation, which would – most likely – have led to the same incriminated products and trace-back results. However, the process would have likely taken considerably longer, as first a trawling questionnaire would have needed to be administered followed by a case-control study.

Trace-back investigation should be undertaken wherever possible to identify the initial source of contamination as this can lead to better source attribution and provide further evidence for policy changes at primary production and processing to reduce the risk of Salmonella entering the food chain. However, it has to be acknowledged that trace-back operations remain cumbersome and time-consuming. During the trace-back operation, the complexity of the food supply chain and diverse stakeholders involved became apparent, once again. All parties involved were inspected by different divisions within the same competent authority and a good mutual communication is essential. Figure 3 gives a peek view of all parties involved in a one-day production of the slaughterhouse. Risks related to contamination at this stage of the supply chain should not be underestimated and can lead to great health risk as well as economic consequences due to recall and destruction of incriminated batches, as shown in this case.

One third of cases involved in the current outbreak was 9 years or younger with the youngest case being two years old. Kivi et al.21 previously showed that younger age (0-9) is a particularly strong risk factor for S. Typhimurium infection (odds ratio 6.3; 95% confidence interval 1.7-23.6), possibly relating to a higher susceptibility of this age group. As the very young, elderly, pregnant and the immunocompromised are known to be at greatest risk of severe disease and death in relation to food- and waterborne infections22, these vulnerable groups should abstain from eating raw or undercooked meat products. Similar to already implemented mandatory provision of allergen information on food product labels23 – we recommend adding explicit warning labels to food products that could potentially be microbiologically unsafe for consumers, especially meat products that are eaten raw or undercooked. As was suggested previously21, adding such information, associated health risks would be transparent and would allow consumers to make well-informed decisions when consuming risk products.

In general, to ensure food safety, batches of products that are intended to be eaten without sufficient heat treatment, should be submitted to risk based HACCP procedures in order to verify compliance with regulations. If such standards are not met – or as a general part of the production process – pre-treating risk products prior to sale could be considered. Pre-treatment methods, such as high-pressure processing that applies pressure to food products and thereby inactivates microorganism without alteration of taste or texture of the food product, constitute promising methods for that purpose24.

In conclusion, this outbreak investigation highlighted the importance of close collaboration between regional Public Health Services, the NVWA, the RIVM and diagnostic laboratories. Through a combined microbiological, epidemiological and source tracing approach, raw and undercooked beef products could be identified as the source of the S. Typhimurium outbreak. This outbreak yet again emphasizes the importance of awareness among consumers of the risk of infection when consuming or handling raw meat products. To increase transparency for consumers, we recommend adding warning labels to risk food products.

Competing Interests

The authors have declared that no competing interests exist.

Data Availability Statement

Access to data are restricted to protect the confidentiality of individuals and premises involved in this outbreak. Researchers interested in accessing an anonymised minimal data set should write to the National Institute for Public Health and the Environment (details below), who will assess the request. Requests for data should be addressed to: Department Epidemiology and Surveillance of Gastroenteritis and Zoonoses (pb 75), Centre for Infectious Disease Control (CIb), National Institute for Public Health and the Environment (RIVM), Pbox 1, 3720 BA Bilthoven, the Netherlands (eelco.franz@rivm.nl).

Corresponding Author

Ingrid Friesema: ingrid.friesema@rivm.nl

]]>
The 2015 Outbreak of Severe Influenza in Kashmir, North India: Emergence of a New Clade of A/H1n1 Influenza Virus http://currents.plos.org/outbreaks/article/the-2015-outbreak-of-severe-influenza-in-kashmir-north-india-emergence-of-a-new-clade-of-ah1n1-influenza-virus-2/ http://currents.plos.org/outbreaks/article/the-2015-outbreak-of-severe-influenza-in-kashmir-north-india-emergence-of-a-new-clade-of-ah1n1-influenza-virus-2/#respond Wed, 08 Aug 2018 13:40:26 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=78750 Introduction: Following the initial outbreak of A/H1N1pdm09, periodic resurgences of the virus, with variable morbidity and mortality, have been reported from various parts of India including the temperate Kashmir region of northern India. An outbreak of A/H1N1 was reported in early 2015 across India with a high morbidity and mortality. We studied patients during the outbreak in Kashmir.

Methods: Patients (n=1780, age 1 month to 90 years, median 35 years) presenting with acute respiratory illness to a tertiary care hospital in Srinagar, Kashmir from October 2014 to April 2015 were recruited. After clinical data recording, combined throat and nasal swabs were collected in viral transport medium and tested by real-time RT-PCR for influenza viruses. All influenza A positive samples were further subtyped using primers and probes for A/H1N1pdm09 and A/H3 whereas influenza B samples were further subtyped into B/Yamagata and B/Victoria lineages. Virus isolation, hemagglutination inhibition testing, sequencing and phylogenetic analysis was carried out using standard procedures. Testing for H275Y mutation was done to determine sensitivity to oseltamivir. All patients received symptomatic therapy and influenza positive patients were administered oseltamivir.

Results: Of the 1780 patients, 540 (30%) required hospitalization and 533 tested positive for influenza [influenza A=517(A/H1N1pdm09=437, A/H3N2=78 with co-infection of both in 2 cases); influenza B=16 (B/Yamgata=15)]. About 14% (n=254) had been vaccinated against influenza, having received the NH 2014-15 vaccine, 27 (11.3%) of these testing positive for influenza.  Sixteen patients, including 4 pregnant females, died due to multi-organ failure. HA sequencing depicted that 2015 isolates belonged to Clade 6B.1. No H275Y mutation was reported from A/H1N1 positives.

Conclusion: Resurgent outbreak of A/H1N1pdm09, with emergence of clade 6B.1, in 2014-15 resulted in high rate of hospitalizations, morbidity and mortality. Periodic resurgences and appearance of mutants emphasize continued surveillance so as to identify newer mutations with potential for outbreaks and severe outcomes.

]]>

Introduction

Following the initial outbreak of A/H1N1pdm09 in 2009-10, periodic resurgences of the pandemic influenza virus have been reported from India with variable morbidity and mortality. 1, 2 We have earlier reported a recrudescence of A/H1N1pdm09 in 2012-13 in Kashmir region of the northern Indian state of Jammu and Kashmir where influenza is an important cause of acute respiratory infections during the winter months. 3, 4

During the initial months of 2015, an unusual increase in influenza A/H1N1pdm09 activity was observed thoughout India with more than 39000 cases and about 3000 deaths. 5 While the activity was usual for the temperate northern state of Jammu and Kashmir, it was unusual for the rest of the country which along with a more tropical geography traditionally witnesses high activity during summers coinciding with the rains of the monsoons. 1 Two scientists from the Massachusetts Institute of Technology, USA in an in silico analysis of retrieved sequences of 2014 from Genbank reported that the higher virulence of the A/H1N1pdm09 in 2015 was attributable to K166Q, D225N and T200A mutations in the HA region of the amino acid sequences in the receptor binding site of A/H1N1pdm09.6 More recently, studies from central, 7 and Eastern India,8 have reported a drift in the A/H1N1pdm09 virus and emergence of different clade of the virus which became the dominant circulating strain in consonance with similar trends across the globe.

We herewith report on the 2015 outbreak in the northern Indian state of Jammu and Kashmir which was associated with high morbidity and mortality and document a change in the genetic constitution of the virus that has implications on the vaccine strain for protection against influenza.

Methods

Kashmir is the major province of the northern most Indian state of Jammu and Kashmir that borders China, Pakistan and Afghanistan. As against the rest of the country with more tropical climate, the valley of Kashmir bound by the Himalayas has a temperate geography with respiratory tract illnesses predominating during the winter months, much like the northern hemispherical seasonality of respiratory viral illnesses seen in Northern America and Western Europe. We have earlier documented pandemic and seasonal influenza viruses as a cause of respiratory illness in Kashmir,3, 4 that constitute the majority of hospital visits during the winter months, either as acute respiratory infections or as infective exacerbations of underlying chronic lung diseases like COPD.9 Sheri-Kashmir Institute of Medical Sciences (SKIMS) is a 820-bed facility in the summer capital, Srinagar, and constitutes the main tertiary referral centre for respiratory cases for the area. During the winter of 2014-2015, SKIMS witnessed an increase in hospital visits by patients with acute respiratory illness, many of whom required hospitalization. We performed surveillance for outpatients with acute respiratory infection (ARI), influenza-like illness (ILI) and in-patients with severe acute respiratory illness (SARI). Influenza like illness (ILI) was defined as fever of 1000F (>37.20C) accompanied by cough and/or sore throat, whereas SARI was defined as those patients with ILI who also require hospitalization. Patients without fever were labeled as ARI.

All patients were interviewed for details of the illness and examined. Clinical history was specifically recorded for any history of contact with a case of ARI or proven influenza and any history of clustering (two or more cases that were related in time and space, e.g., in a home or workplace). After recording of the clinical data, combined throat and nasal swabs were collected in viral transport medium, transported to the influenza laboratory and tested by real-time RT-PCR for influenza viruses using the CDC protocol. 10 Influenza A positive samples were further subtyped using primers and probes for A/H1N1pdm09 and A/H3 and Influenza B positives were subtyped into B/Yamagata and B/Victoria lineages. Virus isolation, haemaggglutination inhibition testing, sequencing and phylogenetic analysis was carried out using standard assay procedures as described previously. 11, 12 Samples were also tested for NA mutations that could result in neuraminidase resistance.

Patients requiring hospitalization were admitted and all patients received symptomatic therapy and influenza positive patients were administered oseltamivir in addition to routine measures that included respiratory support by mechanical ventilation. Post mortem cesarean section was conducted in one deceased pregnant lady for delivering the 34-week fetus that survived after immediate neonatal care for 2 weeks.

In addition to the study samples, at the National Institute of Virology, Pune a total of 285 Indian HA sequences of the period 2009-2015 were rechecked for the 3 mutations reported by Tharakaraman and Sasisekharan. 6 Statistical analysis of all data was done using STATA 11 software using Fisher’s exact test and Chi-square for categorical variables and Student’s t-test for continuous variables. A p value of p<0.05 was considered significant.

Ethics Statement: The study was approved by the Institute Ethics Committee of SKIMS and informed consent for participation was obtained for all patients.

Results

The 1780 recruited patients (845 male; with age 1 month to 90 years (median 35 years) presented with respiratory symptoms of varying severity, 540 (30%) required hospitalisation. The various symptoms experienced by the patients are depicted in table 1 and included respiratory symptoms and fever as the predominant manifestations. The respiratory samples tested positive for influenza in 533 (30%) cases. Table 1 also depicts the distribution of clinical features among influenza positive and influenza negative patients. The median duration of symptoms was 3 days in influenza positive patients compared to a median of 4 days among the influenza negative patients. Influenza positive patients were significantly more likely to have fever, cough, nasal discharge, body aches, fatigue and a history of an acute respiratory tract infection in the family (table 1). Most of the patients presented in the early weeks of 2015 (Figure 1), conforming to the previously documented seasonality of influenza in Kashmir.

Table 1

Demographic and clinical features of influenza-positive and influenza-negative patients.

Influenza positive N (%) Influenza negative N (%) p-value
Number 533 (100) 1247 (100)
Males 262 (49.1) 583 (46.7) 0.35
Age (Mean ± SD) 30.6 ± 19 38 ± 20.5 <0.0001
Duration of symptoms in days (median, range) 3 (1-25) 4 (1-30)
Clinical features
Fever 506 (94.9) 999 (80.1) <0.0001
Cough 503 (94.3) 1098 (88.0) <0.0001
Chills 451 (84.6) 842 (67.5) <0.0001
Nasal discharge 411 (77.1) 956 (76.6) 0.84
Ear discharge 8 (1.5) 22 (1.7) 0.68
Sore throat 388 (72.7) 851 (68.2) 0.05
Breathlessness 351 (65.8) 771 (61.8) 0.10
Expectoration 231 (43.3) 549 (44.0) 0.79
Headache 382 (71.6) 829 (66.4) 0.03
Body ache 407 (76.3) 847 (67.9) 0.003
Fatigue 404 (75.7) 803 (64.3) <0.0001
Concomitant illness 23 (4.3) 121 (9.7) 0.0001
ARI in the family 151 (28.3) 186 (14.9) <0.0001
Vomiting 113 (21.2) 218 (17.4) 0.06
Diarrhea 64 (12.0) 110 (8.8) 0.038
Seizures 8 (1.5) 19 (1.5) 1.0
Vaccination status
Vaccinated 27 (5.0) 227 (18.2) <0.0001

Weekwise positivity of influenza cases (October 2014-March 2015)

Fig. 1: Figure 1.

Weekwise positivity of influenza cases (October 2014-March 2015)

The influenza viruses that were detected in the 533 influenza-positive included 517 cases with influenza A (A/H1N1pdm09=437, A/H3N2=78 with 2 co-infected with both) and 16 cases with influenza B ( B/Yamagata=15)]. HA sequencing and phylogenetic analysis of the A/H1N1 sequences depicted that 2015 isolates from Kashmir clustered with clade 6B.1 with clade specific S84N, S162N and I216T signature mutations (Figure 2). HA-sequencing did not demonstrate any evidence of K166Q, D225 or T200A mutation, even in those who had a fatal outcome of their infection. No H275Y mutation on neuraminidase was reported from pandemic H1N1 positives, hence viruses remained susceptible to oseltamivir.

HA phylogenetic analysis of 2015 isolates of A(H1N1)pdm09 . The tree consists of 2015 isolates from Pune, Delhi and compared with Srinager isolates .The astrix and underline isolates are from the fatal severe cases . The 2015-16 vaccine component is shown in red font. Srinager strains are in blue font. 2015 isolates belong to Clade 7 with clade specific A203T and D97N signature mutations

Fig. 2: Figure 2.

HA phylogenetic analysis of 2015 isolates of A/H1N1pdm09 . The tree consists of 2015 isolates from Pune, Delhi and compared with Srinagar isolates .The asterix and underline isolates are from the fatal severe cases . The 2015-16 vaccine component is shown in red font. Srinagar strains are in blue font. 2015 isolates belong to Clade 6B.1 with clade specific S203T or A203T, and D97N signature mutations

All influenza positive patients received oral oseltamivir, 16 patients died due to multiorgan failure whereas the rest had an uncomplicated course with full recovery. The patients who died included 4 females who had developed influenza during their pregnancy.13

A rechecking of the analysis of the 285 Indian HA sequences of the period 2009-2015 was performed and it was found that K166Q mutation was established in Indian strains from 2013 (Table 2). In addition at position 200, the residue has been ‘A’ itself from the beginning strains of 2009. In 2012 few Indian strains were observed with A200T mutation which did not get established and further it continued as A200 since 2013.

_Revised table for Plos Currents-page0001

Fig. 3: Table 2

Details of the HA sequences of the strains over the years.

Discussion

Our data documents a virulent outbreak of influenza caused predominantly by A/H1N1 virus which upon genotyping belonged to clade 6 as against the classical clade 7 of the A/California09 virus, with few signature mutations associated with the genotype 6.1. There was a high morbidity of the illness with about a third of the patients requiring admission and about 16 influenza-related deaths, including 4 pregnant patients reported earlier. 13 The heightened activity in Kashmir coincided with similar outbreaks in the rest of the country, where previous studies have demonstrated a seasonality coinciding with the summer months and with monsoon rains, 1, 2 even as geographically the country as a whole is geographically located in the northern hemisphere.

Following the emergence of A/H1N1pdm in 2009, the strain largely replaced the circulating seasonal influenza A/H1N1) and Inf A/H3N2 viruses and from 2010 onwards continued to circulate replacing the previously circulating seasonal A/H1N1 along with A/H3 and influenza B. The circulation of these viruses continued with seasonal activity with resurgence of A/H1N1pdm09 in several Indian states in 2012-2013 including the northern state of Jammu & Kashmir.214

Since 2013, several reports have indicated the emergence of an expanding clade of A/H1N1pdm09 viruses, designated 6B. 7, 15, 16 This subgroup appeared in 2012-13 and became predominant in 2013-14. According the World Health Organization, antigenic characteristics of A/H1N1pdm09 viruses collected globally from September 2015 to January 2016 indicated that almost all the A/H1N1pdm09 viruses were antigenically similar and closely related to the vaccine virus A/California/7/2009. These included severe and fatal cases as well. However the sequencing of the HA genes of these viruses indicated the emergence of two new sucbclades within the subgroup 6B (6B.1 and 6B.2) of these A/H1N1pdm09 viruses. The relative proportion of clade 6B viruses expanded from late 2015 and 6B.1 became predominant in most of the geographies except China where the subclade 6B.2 predominated. Our data is in consonance with the global trend observed by the WHO. Pertinently, while the viruses within these subclades are not antigenically distinguishable from A/California/7/2009-like viruses, some recent reports indicate that A/H1N1pdm09 viruses within the 6B.1 and 6B.2 subclades reacted poorly with sera from individuals vaccinated with A/California/7/2009-like-strain-containing vaccine. 17

The first report of genotype 6B strains circulating in India reported by Parida et al7 who demonstrated genotype 6B forming two sub-lineages circulated during the outbreak in Madhya Pradesh in central India harbouring the signature amino acid substitutions of genogroup 6B (D97N, K163Q, S185T, S203T, A256T and K283E). They also noted a new mutation E164G in HA2 sequences. A subsequent study from Eastern India also demonstrated the clustering of the viruses with globally circulating clade 6B. The D225N mutation reported by MIT investigators was not demonstrated but T200A was found to be conserved.8 Our study shows the emergence of subclades of 6B, 6B.1 and 6B.2 in Kashmir, the former predominating.

With a high morbidity and mortality of the 2015 outbreak, it was widely believed that the A/H1N1 virus had mutated and thus rendered more virulent and potentially lethal. The investigators from the Massachusetts Institute of Technology, on the basis of an in-silico analysis of the Genbank submitted Indian-origin strain A/India/6427/2014 reported amino acid T200A and D225N changes that were different from the original A/H1N1pdm09 strain.6 The T200A aminoacid substitution ensures an enhanced human glycan receptor binding of the HA antigen of the influenza virus18 and the D225N substitution leads to increased virulence and disease severity. 19 The D225N mutation has also been reported to affect receptor binding of the HA whereas it also results in reduced susceptibility to neuraminidase (NA), 20 and has previously also been reported to be associated with serious influenza illness requiring hospitalisation or death. 21 The authors believed that the chance of person to person transmission in India with high population density created opportunities for the strain to sustain and they become dominant. The set of mutations that they reported to characterise the strain were K166Q, T200A, and D225N. They also found some N129, G158, and N159 as other important HA changes observed in the retrieved 2014 Indian-origin sequences. Our data along with the analysis of the 285 Indian HA sequences showed that the K166Q mutation was established in Indian strains since 2013. In addition at position 200 the residue has been `A’ itself since the beginning. In 2012 few Indian strains were observed with A200T mutation which did not get established and further it continued as A200 since 2013. The D225N mutation has not been observed in our isolates (including the fatal cases) or other Indian strains with the exception of 3 of 289 Indian strains isolated form severe as well as non severe cases of the year 2013 from Pune. Our isolates clustered with clade 6B.1 and 6B.2. Phylogenetic analysis of other strains from Delhi and Pune also depicted most of the isolates clustering with subclades of clade 6B. Since the circulating strain was drifted from the vaccine strain, it could have resulted in a poorer efficacy of the vaccine in 2014-15. WHO, CDC and the ICMR have now recommended a change of the A/California 09 to the A/Michigan/2015like virus which is likely to have a better match with the circulating A/H1N1virus.

Another important observation in our study was the absence of the H275Y mutation on neuraminidase in our isolates which suggested that the isolates were susceptible to oseltamivir. This has significant implications from the treatment perspective of the patients and as such oseltamivir can continue to be used as per the guidelines.

In conclusion our data demonstrates that A/H1N1 continues to evolve and in the outbreak of 2015 in Kashmir, there was emergence of clade 6B.1 of A/H1N1 which continues to circulate as the dominant strain of seasonal A/H1N1 influenza. Our data emphasize the importance of continued surveillance from wider areas of the country to keep tracking any antigenic changes that would dictate change in the vaccine strain for influenza infection.

Competing Interests

The authors have declared that no competing interests exist.

Data Availability Statement

The data has been uploaded into the public repository, with the DOI as: 10.6084/m9.figshare.5977717.

]]>
http://currents.plos.org/outbreaks/article/the-2015-outbreak-of-severe-influenza-in-kashmir-north-india-emergence-of-a-new-clade-of-ah1n1-influenza-virus-2/feed/ 0
The 2016-2017 Chikungunya Outbreak in Karachi http://currents.plos.org/outbreaks/article/current-chikungunya-outbreak-in-karachi-during-2016/ http://currents.plos.org/outbreaks/article/current-chikungunya-outbreak-in-karachi-during-2016/#respond Tue, 07 Aug 2018 04:20:29 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=80498 Introduction: Chikungunya is an incipient disease, caused by Chikungunya virus (CHKV) that belongs to genus alphavirus of the family Togaviridae.

Materials and Methods: In this study, during an outbreak of CHKV in Dec 2016 in Karachi, Pakistan, samples were collected from patients presenting with fever, tiredness and pain in muscles and joints. Total 126 sera were tested for the presence of Chikungunya infection through ELISA and Real-time Reverse Transcriptase PCR assay.

Results and Discussion: This study showed that approx 79.4% samples were positive for CHKV. To our knowledge, this is the first reported outbreak from last decades in which the presence of CHKV is confirmed in Karachi while affecting such large no. of individuals.. Conclusion: CHKV diagnosis should be considered by the scientists and clinicians as a differential diagnosis in febrile patients, and appropriate control strategies must be adopted for its surveillance.

]]>

Introduction

Chikungunya is a fever causing contagion which is transmitted to humans through the bite of CHKV harboring mosquitoes A. aegypti and A. Albopictus. The name “chikungunya” is derived from a word of the Kimakonde language which means “to become contorted”, and describes the stooped appearance of sufferers with joint pain (arthralgia). CHKV generally causes mild illness but could lead to severe life-threatening complications. The disease is characterized by an acute illness with fever, chills, headache, nausea, vomiting, joint pain, low back pain, and skin rash. CHKV causes arthralgia, which may persist for months 1, 2, 3 . The incubation period of CHKV ranges from 2–10 days, with statutory symptoms lasting up to 7 days. The symptoms usually resolve within days to a few weeks; but in severe cases, these symptoms may last for months. Currently, chikungunya fever has affected more than 50 countries. The global distribution of A. aegypti is expanding owing to global travel and trade, and so does the virus. It has become a public health problem in Asia, Africa, Europe and America. In 1983 chikungunya existence was reported in rodents in Pakistan. Until 2007 Pakistan was not included in CDC list. In 2008 Pakistan appeared in the list and cases were first reported in 2011 when dengue cases were atpeak in the country. During Dengue fever outbreak in 2011 a few patients with chikungunya were reported in Lahore 4 . Current outbreak of 2016-17 was initially termed as Mysterious disease 5 associated with the warm climate and inferior sanitary state of the city 4 . According to a local report total suspected cases between Dec 19, 2016, and Feb 22, 2017 were 818 6 , and according to WHO a total of 1018 suspected cases of chikungunya were reported between 19 December 2016 to 30 March 2017, in various districts of Karachi. No deaths have been reported so far 7 . These cases were clinically evaluated at various hospitals and labs in Karachi and Islamabad. At NIH, Islamabad 121 samples out of 157 samples were confirmed for CHKV infection via laboratory diagnosis 7 . In this study, total 126 serum samples from febrile illness patients were evaluated at CESAT to confirm an outbreak of CHKV infection in Karachi for the first time.

Materials and Methods

During December 2016, clinicians in Karachi (Pakistan) observed more than 1000 patients (in a single day) presenting with fever, body rash, swollen joints and joint pain in three large hospitals of Karachi at the area of Saudabad, Malir and its surrounding.

Sample Collection: Total 126 blood samples were mainly collected from suspected patients admitted in emergency of Saudabad Sindh Govt Hospital, Karachi. These patients were presented with fever, chills, body pain, headache, arthralgia and anorexia. Most of the patients had stooped figure due to pain. Blood samples were collected in gel tubes and EDTA tubes for ELISA and PCR respectively.

Demographics and Travel History: Total 68 male and 58 female patients were included in this study. None of the patients had history of traveling abroad prior to infection. Patients were mostly from Malir Area of Karachi which is a mosquito endemic area, with low hygienic conditions.

Selection Criteria: The inclusion criteria for the patient were high grade fever, severe joint pain, stiff hands and stooped figure.

ELISA: Serum was separated from blood samples and all the serum samples were screened for the presence of Immunoglobulin M (IgM) antibodies against CHKV using a microtiter plate ELISA assay. The ELISA test was performed according to the kit manufacture’s protocol and interpreted either positive or negative on the basis of absorbance with respect to cutoff values.

PCR: In the present study, RNA samples were also analyzed for molecular detection by using Real-time Reverse Transcriptase PCR. Chikungunya infections results in high levels of viremia, which typically last for 4–6 days after the onset of illness. During the acute phase of infection viral RNA can be easily detected by reverse transcriptase-polymerase chain reaction (RT-PCR) in serum samples obtained from patients. RT-PCR can therefore easily be done within the first 7 days on an acute-phase specimen to confirm chikungunya virus infection. RNA was extracted from 400μl of serum samples. One-step Real time PCR assay was carried out using Primer pair and Probes specific to CHKV. In singleplex reaction mixture 11μl of RNA template was added with 14 μl of PCR Master mix (Invitrogen), to make 25 μl total.

Ethics Statement: Samples were collected after receiving approval from the local hospital Ethics Committee. Obtained written informed consent was received from the participants.

Results & Discussion

The CHKV and DENV vectors i.e. Ae. Aegypti and Ae. Albopictus, already exist and thrive in Pakistan. Therefore, both Chikungunya and Dengue virus have opportunity of infecting the community on large scale. Furthermore, the initial signs and symptoms of both Dengue and Chikungunya are quite similar, which may lead to difficulties in making an appropriate provisional diagnosis. Laboratory diagnosis by ELISA and PCR both plays a vital role for differential diagnosis between Dengue and Chikungunya.

Out of total of 126 suspected serum, thirty eight (38) samples were positive for anti-Chikungunya IgM by ELISA, which suggest that these patients were in post viremic or convalescent phase. Three (03) patients were found positive for anti-Chikungunya IgG by ELISA alone which suggest that these patients were in post viremic or late stage after infection. Forty nine (49) samples were found positive by Real-Time Reverse Transcriptase PCR assay alone, which suggest that these patients were in viremic or acute phase. However, eighteen (18) samples were found positive by ELISA and PCR both, which suggest that these patients were in transitional phase.

This study revealed that overall 100 (79.4%) patients were found affected out of 126 suspected patients during this outbreak. This is a huge number of affected populace in a single city at a time.

Conclusion

This study confirmed that the current outbreak in Karachi during Dec 2016-Feb 2017 is of Chikungunya Virus and may appeared in more frequent outbreaks of CHKV in near future as of dengue in Pakistan.

We recommend surveillance for CHKV, its vectors and preparedness to prevent future outbreaks of CHKV infection in Pakistan.

Corresponding Author

Dr. Nadia Jamil is the corresponding author for this article. Email: diaraj@hotmail.com

Data Availability Statement

The raw data required to reproduce these findings are available to download from the following Figshare link: https://figshare.com/articles/Probes_and_Primers_details/6917054.

Example photographs can be found via the following Figshare DOIs: 10.6084/m9.figshare.6819419; 10.6084/m9.figshare.6819431; 10.6084/m9.figshare.6819434; 10.6084/m9.figshare.6819437; 10.6084/m9.figshare.6819440; 10.6084/m9.figshare.6819443.

Competing Interests Statement

There are no competing interests.

]]>
http://currents.plos.org/outbreaks/article/current-chikungunya-outbreak-in-karachi-during-2016/feed/ 0
Identifying Areas at Greatest Risk for Recent Zika Virus Importation — New York City, 2016 http://currents.plos.org/outbreaks/article/identifying-areas-at-greatest-risk-for-zika-virus-importation-new-york-city-2016/ http://currents.plos.org/outbreaks/article/identifying-areas-at-greatest-risk-for-zika-virus-importation-new-york-city-2016/#respond Wed, 25 Jul 2018 14:20:08 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=76289 Introduction: The New York City Department of Health and Mental Hygiene sought to detect and minimize the risk of local, mosquito-borne Zika virus (ZIKV) transmission. We modeled areas at greatest risk for recent ZIKV importation, in the context of spatially biased ZIKV case ascertainment and no data on the local spatial distribution of persons arriving from ZIKV-affected countries.

Methods: For each of 14 weeks during June-September 2016, we used logistic regression to model the census tract-level presence of any ZIKV cases in the prior month, using eight covariates from static sociodemographic census data and the latest surveillance data, restricting to census tracts with any ZIKV testing in the prior month. To assess whether the model discriminated better than random between census tracts with and without recent cases, we compared the area under the receiver operating characteristic (ROC) curve for each week's fitted model versus an intercept-only model applied to cross-validated data. For weeks where the ROC contrast test was significant at P < 0.05, we output and mapped the model-predicted individual probabilities for all census tracts, including those with no recent testing.

Results: The ROC contrast test was significant for 8 of 14 weekly analyses. No covariates were consistently associated with the presence of recent cases. Modeled risk areas fluctuated across these 8 weeks, with Spearman correlation coefficients ranging from 0.30 to 0.93, all P < 0.0001. Areas in the Bronx and upper Manhattan were in the highest risk decile as of late June, while as of late August, the greatest risk shifted to eastern Brooklyn.

Conclusion: We used observable characteristics of areas with recent, known travel-associated ZIKV cases to identify similar areas with no observed cases that might also be at-risk each week. Findings were used to target public education and Aedes spp. mosquito surveillance and control. These methods are applicable to other conditions for which biased case ascertainment is suspected and knowledge of how cases are geographically distributed is important for targeting public health activities.

]]>

Introduction

The New York City (NYC) population, which includes many travelers and recent immigrants, is at risk for travel-associated communicable diseases, including Zika virus (ZIKV) disease.1,2,3 Infected persons who acquire ZIKV while traveling and are viremic in NYC might be bitten by Aedes albopictus vectors, introducing a risk of local, mosquito-borne transmission. Any such locally-acquired cases might not be easily recognized, as an estimated 80% of ZIKV infections are asymptomatic,4 clinical symptoms when present can be mild and non-specific, and healthcare providers would need to suspect local transmission to order ZIKV testing for patients without a relevant travel history.5

The risk of local, mosquito-borne ZIKV transmission in NYC during 2016 was considered low because the primary vector species, Aedes aegypti, is not present in NYC. Nevertheless, given limited evidence that A. albopictus could transmit ZIKV,6,7 a high population density, large numbers of potentially infected travelers, and the serious health consequences of congenital ZIKV infection including microcephaly,8 the NYC Department of Health and Mental Hygiene (DOHMH) was concerned about local transmission and aimed to detect such an unlikely occurrence as early as possible to institute measures to interrupt transmission to humans. DOHMH conducted intensive mosquito surveillance to assess the abundance of the mosquito vector and to test mosquitoes for ZIKV.

In parallel, DOHMH also sought to identify locations with higher numbers of persons with ZIKV viremia, as these would be areas where mosquitoes could become infected with ZIKV. There were several challenges. First, the geographic distribution of NYC residents at risk for travel-associated ZIKV infection is unknown. While data on the volume of commercial air passenger arrivals are available by airport,9,10 no data are readily available on where or whether these travelers reside within NYC. Second, while acute arboviral infections, including ZIKV infections, are reportable by healthcare providers and laboratories to NYC DOHMH,11 these data are incomplete. Recently reported cases are likely a spatially non-representative sample of all currently viremic persons, given testing and reporting lags and the high proportion of asymptomatic and subclinical ZIKV infections.4 Furthermore, not all at-risk NYC patients were equally likely to receive ZIKV testing, given initial barriers to seeking care and receiving testing for ZIKV in areas with high poverty and large numbers of persons born in countries with local ZIKV transmission.3,12 Many cases are detected after viremia has ended and are only diagnosed serologically. Thus, mapping the distribution of the residences of reported cases with recent diagnoses can be useful13 but is not necessarily sufficient to characterize all areas at highest risk of ZIKV importation. Third, at-risk areas need to be dynamically updated throughout the mosquito-borne disease transmission season to reallocate resources and target new areas as needed. We expected the geographic distribution within NYC of imported ZIKV cases could change over time, reflecting changing patterns of risk to populations living in different areas of NYC, caused by changes in incidence and outbreak dynamics across multiple ZIKV-affected countries.14

Given that reported ZIKV cases were incomplete and spatially non-representative, we aimed to identify additional areas in NYC potentially at risk for ZIKV importation by using area-level, static census data and the latest data available to DOHMH on ZIKV testing. In this study, we used logistic regression15 to predict weekly nowcasts16 throughout the summer of 2016 of census tracts at greatest risk of recent ZIKV importation. Nowcast results were used to inform geographically targeted activities, including performing public education, enrolling additional healthcare facilities in a sentinel surveillance system for detecting local ZIKV transmission, interpreting syndromic surveillance signals suggesting possible ZIKV-like illness, and, when reviewed in conjunction with mosquito surveillance data, informing control of Aedes spp. mosquitoes and placement of traps for continued surveillance.3,17

Methods

Data Sources

The NYC population (an estimated >8.5 million persons as of July 2015)18 was eligible for analysis. The unit of analysis was 2010 census tract (n=2,123 in NYC with >25 residents), i.e., the finest geographic resolution available for all independent variables. We selected small geographic units to prioritize spatial precision in identifying areas at high risk, despite potential instability in estimates for some geographic units. Smaller units have more homogeneous risk factor distributions than larger units, minimizing inferential problems in ecologic analysis.19 Census tract-level sociodemographic data were obtained from the 2010 U.S. Census and the American Community Survey 2010–2014. De-identified ZIKV-related testing and case data were obtained from the disease surveillance database used by the DOHMH Bureau of Communicable Disease (Maven, Conduent Public Health Solutions, Austin, TX) and aggregated to census tract resolution prior to analysis.

Variables

The dependent variable was the census tract-level presence of any recent ZIKV cases reported to DOHMH. Cases included persons with confirmed or probable laboratory evidence of ZIKV infection or disease20 and persons who tested positive for ZIKV by IgM with pending plaque reduction neutralization test results. A recent case was defined as having an “event date” (illness onset date if available, otherwise specimen collection date) in the 28 days prior to the weekly data extract. A 28-day period was selected to approximately encompass the risk of an imported ZIKV case leading to A. albopictus being currently or imminently infectious as of each analysis, considering a 7-day period of viremia in a human post-illness onset and a 14–20-day period before a mosquito becomes infectious.

The input dataset also contained eight independent variables; data for six of these variables were static. We included census tract-level sociodemographic characteristics that we suspected might be associated with the underlying risk of ZIKV importation and/or testing. As the distribution of persons at risk for travel-associated ZIKV infection is unknown, the first and second static variables were used as proxies for the number of travelers from selected areas with local mosquito-borne ZIKV transmission (Mexico, the Caribbean, Central America, and countries in South America):21 the number of persons born in these countries [1], grouped by quartile; and the number of persons with ancestry from these countries [2], grouped by quartile. Third, since travel-associated communicable disease incidence is consistently associated with area-based poverty in NYC,22 we included the proportion of the population living below the federal poverty level. Data for those first three variables were from the American Community Survey 2010–2014. We also included the proportion of the population of Hispanic ethnicity, the proportion of the population of women of childbearing age (15–44 years-old), and the quartile of the total population size, per the 2010 Census.

Data for the remaining two independent variables were updated weekly to reflect the most current available ZIKV-related data in the disease surveillance database. Since historical cases might be predictive of more recent cases, we included the number of ZIKV cases with an event date >28 days prior to the weekly data extract. Finally, since ZIKV, dengue virus, and chikungunya virus disease each occur among persons traveling from similar geographic areas, we included the cumulative number of confirmed and probable dengue and chikungunya cases since January 2013 who reported on interview from routine case investigation to have traveled to selected countries [3] with local mosquito-borne ZIKV transmission.21

An additional indicator variable, whether any ZIKV tests were ordered for residents of each census tract in the prior month, was updated weekly and used not as an independent variable, but for restriction in model fitting, as described in the next section. For most notifiable diseases, only positive laboratory results are reportable by law. Access to testing data was unusual but possible in this situation because ZIKV testing was coordinated by DOHMH and performed by public health laboratories before commercial testing became available.3,23 Time and climatic factors were not included in the model, since our objective was to use the most recent available data to answer a purely spatial (not explicitly spatio-temporal) question, i.e., identifying areas at greatest risk for recent ZIKV importation as of each weekly analysis.

Weekly Nowcasting Statistical Analysis

Preliminary analyses were performed May–June 2016, and 14 weekly analyses were conducted throughout the peak mosquito season for comparison with vector surveillance data, from the week starting June 26 through the week starting September 25, 2016. Each week, we used multivariable logistic regression using Firth’s penalized maximum likelihood estimation to model the presence of any recent ZIKV cases, given all eight covariates described above. We used regression modeling rather than spatial interpolation or smoothing methods because we did not want to inappropriately smooth results across adjacent neighborhoods with heterogeneous sociodemographic characteristics and travel patterns. We fit the model restricting to census tracts with any recent testing, i.e., excluding census tracts where having zero observed recent cases could be attributable to no testing.

Each week, we output the census tract-level leave-one-out cross-validated predicted probabilities24 of any imported ZIKV cases in the prior month, using the OUTPUT statement in SAS (predprobs=crossvalidate, SAS Institute, Cary, NC; see Supporting Information). In brief, the data for any given census tract were omitted, the logistic regression model was fit using data from all other census tracts, and then the predicted probability for the given census tract was estimated using the observed values for the given census tract and the modeled parameter estimates. To account for geographic variation in testing uptake, we required census tracts to have any ZIKV testing in the prior month to be included in model fitting. Model performance was assessed using receiver operating characteristic (ROC) curves, which determined the discriminatory accuracy of each week’s model to identify census tracts with any recent imported ZIKV cases. An ROC contrast test was used to compare the area under the ROC curve (AUC) for the fitted model compared with an intercept-only model (i.e., random prediction, AUC=0.5) applied to the cross-validated data.25 If the P-value from the ROC contrast test was <0.05, then we proceeded to output the individual predicted probabilities of any recent imported ZIKV cases (the “nowcast”) for all census tracts, including those with no recent testing.

We refined this modeling process during and after the mosquito season; thus, we present results of the finalized modeling process as retrospectively applied to weekly archived data extracts. These data extracts represented the data actually available to DOHMH in real-time and were not only incomplete because of reporting lags, but also preliminary because data cleaning was not yet completed. All observations (census tracts) were considered to be independent, ignoring spatial autocorrelation, because we needed unbiased point estimates but were not concerned with the variance of those estimates.26

Any covariates significantly associated with the outcome each week according to the type III analysis of effects were noted. We assessed variation in nowcast results over time using Spearman rank correlation coefficients for the modeled ZIKV importation risk per census tract. In this analysis, the rank values for the census tract-level predicted probabilities of any recent imported ZIKV cases were compared between weeks. A Spearman rank correlation coefficient value of 1 would indicate no changes in the rankings between weeks (such that the same census tracts were always at greatest risk), a value of 0 would indicate no association in the ranks, and a value of -1 would indicate a perfect negative association of ranks. Census tracts in the highest decile of modeled ZIKV importation risk were visualized (ArcMap 10.2.1, Esri, Redlands, CA).

Ethics Statement

This activity involved the use of data collected for non-research purposes, and there were no interactions or interventions with living individuals. All efforts were made to protect individual privacy and anonymity. Data were de-identified before being accessed and used for the purpose of this activity. The scope of this activity was limited to public health practice, and all activities were authorized and conducted by NYC DOHMH, a public health authority that is responsible for such public health matters as part of its official mandate. This activity was categorized as public health surveillance by the NYC DOHMH’s Institutional Review Board.

Results

During June–September 2016, 652 cases of ZIKV infection or disease were observed among residents of 475 census tracts in NYC. All reported cases were travel-associated, including travelers returning from ZIKV-affected areas, their sexual contacts, and infants infected in utero. Zero cases were observed in 1,648 (78%) of the 2,123 census tracts with >25 residents. Across the 14 weekly analyses, the median number of census tracts with any recent ZIKV testing and thus used to fit the weekly models was 688 (range: 516–827), and of these, the median number of census tracts with any recent ZIKV cases was 102.5 (range: 51–120).

Across the 14 weekly analyses, no covariates were consistently significantly associated with the presence of recent ZIKV cases (Table 1). One covariate (the proportion of the population of women of childbearing age) was significantly associated with ZIKV cases during 3 of the analysis weeks. Five covariates were each significantly associated with the outcome during 2 analysis weeks, 1 covariate was significantly associated during 1 analysis week, and 1 covariate was never significantly associated. During 4 weekly analyses (July 29, August 3, September 7, and September 13), no covariates were significantly associated with the outcome. Toward the beginning of the peak mosquito season, the covariates most strongly associated with recent ZIKV cases were historical counts of dengue, chikungunya, and ZIKV cases. By August, demographic factors (ethnicity, sex and age, and ancestry) instead were most strongly associated. By late September, only poverty level was significantly associated (Table 1).

Table 1. Variables statistically significantly associated with recent Zika virus importation and model performance characteristics, by week of analysis, New York City, 2016

Table 1: Variables statistically significantly associated with recent Zika virus importation and model performance characteristics, by week of analysis, New York City, 2016

According to the ROC contrast test, the model discriminated better than random between census tracts with and without recent cases for 8 of 14 weekly analyses (Table 1). In these weeks, the AUC ranged from 0.57 to 0.66, where AUC=0.5 represents random prediction and AUC=1 represents perfect prediction.

We proceeded to output nowcasts for these 8 weekly analyses, all of which were strongly correlated (P <0.0001). However, variation in the modeled ZIKV importation risk per census tract was observed, with Spearman correlation coefficients ranging from 0.30 (indicating a weak positive association in the ranks of census tracts at risk between August 30 and September 20) to 0.93 (indicating a strong positive association in the ranks of census tracts at risk between August 16 and August 23, Table 2). The variation in Spearman correlation coefficients indicated shifts in areas at risk over time. For example, as of June 29, census tracts in the highest decile of modeled recent ZIKV importation risk were concentrated in the Bronx and upper Manhattan (Fig 1), while as of August 23, the greatest risk had largely shifted to eastern Brooklyn (Fig 1). In one defined Brooklyn neighborhood tabulation area consisting of 33 census tracts, 23 census tracts were in the highest decile of modeled risk as of August 23, but only 10 census tracts had any recent testing, and only 3 census tracts had observed recent cases.

Correlations* of nowcasts for census tract-level risk of recent Zika virus importation as of eight time points with adequate model performance, New York City, 2016

Table 2: Correlations* of nowcasts for census tract-level risk of recent Zika virus importation as of eight time points, New York City, 2016

Census tracts in the highest decile of modeled risk of Zika virus importation as of nowcast for four time points, New York City, 2016

Fig. 1: Census tracts in the highest decile of modeled risk of Zika virus importation as of nowcast for four time points, New York City, 2016.

Discussion

Our dynamic process used the most recent available data to nowcast how imported ZIKV cases were likely to be currently distributed in near real-time. While most published efforts to identify areas at risk of ZIKV importation have been at the country-level,27,28 local and state health departments with persons at risk for travel-associated diseases require more geographically refined estimates to inform programmatic activities and to detect and mitigate local transmission.29

By mapping the nowcast output, we visualized shifts in the spatial distribution of risk over time. For instance, as of late June, the greatest risk of recent ZIKV importation was concentrated in the Bronx and upper Manhattan, and the only covariate significantly associated with risk was historical counts of chikungunya and dengue cases, which had been similarly concentrated in these areas.30 In contrast, by late August, the area at greatest risk had largely shifted to eastern Brooklyn, and risk was instead associated with specific population demographic factors, consistent with targeted interventions to increase ZIKV testing at facilities serving this population.12,31 Such shifts might reflect changing outbreak dynamics;14 i.e., as the force of infection peaked in different source countries at different times, different subpopulations of arriving travelers living in geographically distinct communities were therefore affected at different times, changing the geographic distribution of imported ZIKV cases in different parts of NYC. By definition of having a model with better than random discriminatory ability, we expected the nowcast to identify neighborhoods with recent observed cases. We noted situations where the nowcast also identified a neighborhood with a high proportion of at-risk census tracts but only a few observed cases. We also noted situations where the nowcast identified areas (groupings of census tracts) at-risk but with no recent observed cases, as this could inform public health operations, including targeting public education and setting traps for and controlling Aedes spp. mosquitoes.

Because ZIKV testing had been arranged primarily by DOHMH in 2016, we were able to restrict the model fitting to census tracts with recent testing. However, with most testing in 2017 done commercially and no access to overall testing data, such data will not be readily available going forward. Public health authorities might consider alternative methods of obtaining geographically resolved testing data, e.g., by partnering with commercial laboratories or regional health information organizations, or by making negative test results reportable. For example, in 2014, the NYC Health Code was amended to mandate reporting of negative hepatitis C nucleic acid test results.32 Even if testing data are not available, this method could still be useful in identifying areas where cases are not ascertained for reasons other than testing biases, e.g., because infections are asymptomatic and not medically attended.

This work is subject to at least three limitations. First, while internal cross-validation determined that the model predicted better than random for 8 of 14 analysis weeks, the maximum AUC across weeks was only 0.66. Model performance would likely improve if census tract-level data on persons arriving in NYC from ZIKV-affected countries were readily available. We displayed census tracts in the highest decile of modeled ZIKV importation risk as of a given week, although this threshold is arbitrary. It is unknown whether census tracts classified as high risk truly had any cases, as external validation would have required special outreach to test persons who would not otherwise have sought ZIKV testing and was not logistically feasible. Nowcasts might also be improved by excluding persons whose viremia resolved prior to arriving in NYC from a ZIKV-affected country. Second, analyses were based on patient residence and did not account for other locations in NYC where patients might have spent time while viremic. Third, the results were subject to the ecologic fallacy, such that inferences drawn from observing census tracts might not necessarily apply to all census tract residents.19 Nevertheless, the nowcasts were useful for targeting resources geographically.

The weekly nowcasts of modeled ZIKV importation risk were overlaid on vector surveillance data regarding mosquito abundance to improve situational awareness for DOHMH leadership and the ability to target preventive measures. No local ZIKV transmission was observed in NYC in 2016. These methods could be applied to other conditions for which biased case ascertainment is suspected and understanding the full geographic distribution of cases at any moment is important for targeting public health activities. Future work can include exploring complementary approaches for nowcasting and forecasting ZIKV importation and local transmission risk at high geographic resolution, e.g., by using agent-based models.33

[1] Bahamas, Barbados, Belize, Bolivia, Brazil, Colombia, Costa Rica, Cuba, Dominica, Dominican Republic, Ecuador, El Salvador, Grenada, Guatemala, Guyana, Haiti, Honduras, Jamaica, Mexico, Nicaragua, Panama, Puerto Rico, Peru, St. Vincent and the Grenadines, Trinidad and Tobago, U.S. Island Areas, Venezuela, West Indies, and Other Caribbean. From ACS 2010–2014 tables B05006: place of birth for the foreign-born population in the United States, and B05002: place of birth by nativity and citizenship status. Countries not specified in these tables include: Aruba, Bonaire, Curacao, French Guiana, Guadeloupe, Martinique, Paraguay, Saint Barthélemy, Saint Lucia, Saint Martin, Sint Maarten, and Suriname.

[2] Bolivian, Brazilian, Colombian, Costa Rican, Cuban, Dominican, Ecuadorian, Guatemalan, Guyanese, Honduran, Mexican, Nicaraguan, Panamanian, Paraguayan, Peruvian, Puerto Rican, Salvadoran, Venezuelan, West Indian. From ACS 2010–2014 tables B04006: people reporting ancestry, and B03001: Hispanic or Latino origin by specific origin. Not specified in these tables include persons with ancestry from: Aruba, Bonaire, Curacao, French Guiana, Guadeloupe, Martinique, Saint Lucia, Saint Martin, Saint Vincent and the Grenadines, Sint Maarten, and Suriname.

[3] Aruba, Anguilla, Antigua and Barbuda, Barbados, Bahamas, Belize, Bolivia, Bonaire, Saint Eustatius and Saba, Brazil, British Virgin Islands, Cayman Islands, Colombia, Puerto Rico, Costa Rica, Cuba, Curacao, Dominica, Dominican Republic, Ecuador, El Salvador, French Guiana, Grenada, Guadeloupe, Guatemala, Guyana, Haiti, Honduras, Jamaica, Montserrat, Turks and Caicos Islands, Martinique, Mexico, Nicaragua, Panama, Paraguay, Peru, Saint Barthelemy, Saint Kitts and Nevis, Saint Lucia, Saint Martin, Saint Maarten, Saint Vincent and the Grenadines, Suriname, Trinidad and Tobago, U.S. Virgin Islands, and Venezuela.

Supporting Information

SAS code to generate the weekly nowcast of modeled risk of Zika virus importation and assess model performance is available here: https://github.com/CityOfNewYork/communicable-disease-surveillance-nycdohmh

Competing Interests

The authors have declared that no competing interests exist.

Funding

S.K.G. was supported by the Public Health Emergency Preparedness Cooperative Agreement (grant NU90TP000546) from the Centers for Disease Control and Prevention. S.L. and A.D.F. were supported by New York City tax levy funds. The contents of this publication are solely the responsibility of the authors and do not necessarily represent the official views of the Centers for Disease Control and Prevention or the Department of Health and Human Services. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Data Availability

Census tract-level sociodemographic data were obtained from the 2010 U.S. Census and the American Community Survey 2010–2014 and are available from the U.S. Census website (https://www.census.gov/data.html). Census tract-level ZIKV testing and case data are not publically available from DOHMH in accordance with patient confidentiality and privacy laws; such data can be made available by contacting BCD_reportable_data@health.nyc.gov and obtaining institutional review board approval and executing a data use agreement approved by the legal departments of participating institutions.

]]>
http://currents.plos.org/outbreaks/article/identifying-areas-at-greatest-risk-for-zika-virus-importation-new-york-city-2016/feed/ 0
Lessons Learnt from Epidemiological Investigation of Lassa Fever Outbreak in a Southwest State of Nigeria December 2015 to April 2016 http://currents.plos.org/outbreaks/article/lessons-learnt-from-epidemiological-investigation-of-lassa-fever-outbreak-in-a-southwest-state-of-nigeria-december-2015-to-april-2016/ http://currents.plos.org/outbreaks/article/lessons-learnt-from-epidemiological-investigation-of-lassa-fever-outbreak-in-a-southwest-state-of-nigeria-december-2015-to-april-2016/#respond Fri, 29 Jun 2018 11:45:58 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=77487 Introduction: An outbreak of Lassa Fever (LF) reported and confirmed in Ondo state, Southwest Nigeria in January 2016 was investigated. This paper provides the epidemiology of the LF and lessons learnt from the investigation of the outbreak.

Methods: The incidence management system (IMS) model was used for the outbreak coordination. Cases and deaths were identified through the routine surveillance system using standard definitions for suspected and confirmed cases and deaths respectively. Blood specimens collected from suspect cases were sent for confirmation at a WHO accredited laboratory. Active case search was intensified, and identified contacts of confirmed cases were followed up for the maximum incubation period of the disease. Other public health responses included infection prevention and control, communication and advocacy as well as case management. Data collected were analysed using SPSS 20, by time, place and persons and important lessons drawn were discussed.  

Results: We identified 90 suspected LF cases of which 19 were confirmed by the laboratory. More than half (52.6%) of the confirmed cases were females with majority (73.7%) in the age group ≥ 15 years. The Case Fatality Rate (CFR) of 63.2% among the laboratory-confirmed positive cases where 9 of 19 cases died, was significantly higher compared to the laboratory confirmed negative cases where 6 of the 65 cases died ( CFR; 8.5%) p ≤ 0.05. Two hundred and eighty-seven contacts of the confirmed cases were identified, out of which 267(93.0%) completed  the follow-up without developing any symptoms and 2 (0.7%) developed symptoms consistent with LF and were confirmed by the laboratory. More than half of the contacts were females (64.5%) with most of them (89.2%) in the age group ≥ 25 years.  

Discussion: One key lesson learnt from the investigation was that the confirmed cases were mainly primary cases; hence the needs to focus on measures of breaking the chain of transmission in the animal-man interphase during Lassa fever epidemic preparedness and response. In addition, the high case fatality rate despite early reporting and investigation suggested the need for a review of the case management policy and structure in the State. Key Words: Lassa fever, Outbreak Response, Incident Management System, Nigeria

]]>

Introduction

Lassa fever (LF) is a severe acute viral hemorrhagic illness caused by a virus belonging to the family Arenaviridae.1,2 The disease was first discovered in Sierra Leone in the 1950s, but the aetiological agent was first isolated after an outbreak of the disease in a village called Lassa in Borno State, Nigeria claiming the lives of two foreign missionary nurses in 1969.3

The virus exhibits persistent, asymptomatic infection, with profuse urinary virus excretion in Mastomys natalensis rodents, which serve as the natural reservoir.1,2 The virus is shed in their excreta (urine and faeces) of the rodent which can be aerosolized and inhaled by humans.4 Primary mode of spread is from rodent to man through contact with rodent excreta or urine in food or during hunting and processing of rats for consumption. The virus can spread from person-to-person, either within households during care for sick relatives or in health care settings.5

The disease, LF, is endemic in West Africa with several outbreaks recorded over the years. Outbreaks of the disease have been in reported in Sierra Leone, Guinea, Liberia, Nigeria, Ghana, and Ivory Coast, Senegal and Mali.6 The number of infections per annum has been estimated at 100,000 to 300,000 with approximately 5,000 deaths.6,7,8,9,10 Since the identification of the virus in Nigeria in 1969, yearly outbreaks have been reported in parts of the country,11,12 and more recently in some states including Ondo States.13,14,15,16

Despite the yearly outbreaks of LF reported in the country, few literatures exist on the detailed epidemiological investigation and coordinated public health responses used for the control of these outbreaks, especially with modern approaches such as the Emergency Operation Center (EOC) model and most importantly, documented lessons learnt for improved future outbreak responses.

Between December 2015 and February 2016, Nigeria reported suspected cases of LF cases from about 23 states, including Ondo State with preliminary epidemiological and laboratory investigation confirming an outbreak of the disease. The need to document key lessons learnt from the outbreak investigation in Ondo State, in order to improve future responses, incentivises this report. This paper, therefore, describe the epidemiology of the LF outbreak in the State, outline public health responses conducted and document some key lessons learnt to enhance outbreak response.

Methods

Outbreak setting

Nigeria is the most populous country in Africa, with an estimated population of over 160 million (Samson, 2014) and a growth rate of 3.8% per annum. Nigeria has six regional zones with varying ecologies, climates and population characteristics. The zones are divided into 36 states and the federal capital territory, which is further divided into 774 LGAs or districts and 8 812 administrative wards.17 Ondo State is one of the 36 states in the Federal Republic of Nigeria situated between longitudes 40 151 E and 60 001E of the Greenwich meridian and latitudes 50 451N and 70 451 N, which are to the North of the equator, in the Southwestern geopolitical zone of the country.17

The State has 18 LGAs with three senatorial districts; Ondo North, Central and South and a 2015 projected total population of about 4,489,756 based on the 2006 population census.18 The outbreak was restricted to eight LGAs in the north and central senatorial districts of the state. The climate of the areas is highly favoured for the agrarian activities and crops such as cocoa, kola nut, palm tree and arable crops like maize and tubers such as yam and cassava are grown annually.19 The annual rainfall is between 1000mm and 1500mm with a high daily temperature of about 300C. The vast majority of the population consists of peasant farmers cultivating food and cash crops at a small-scale level. Livestock keeping is a minor occupation of the population of Ondo State who rear goats, sheep and also do some fish farming. Other economic activities in the state include trading and civil service.20

Field Investigations and public health response

Following an alert from the disease surveillance and notification officer of two suspected cases of LF at the State Specialist Hospital, Akure on 8 January 2015, the Ondo State Ministry of Health (SMoH) through her Epidemic Management Committee and the World Health Organization (WHO) Ondo State field office reactivated the existing Incident Management System (IMS) at the SMoH, Akure modeled after the Emergency Operating Centre (EOC) for the control of Ebola Virus Disease (EVD) in Nigeria.21 The Incident Management System (Appendix I) has previously been used in the investigation of methanol poisoning in the state in 2015.22 Strategic groups, which included Epidemiology and Surveillance committee, Case management committee, Infection Control and Burial committee as well as Advocacy and Communication committee were inaugurated and served as the implementation units of the State outbreak response plan for the control the outbreak (Appendix I) .

Coordination

The investigation of the outbreak as well as implementation of the outbreak response plan was coordinated by the SMoH officials with technical support provided by the WHO Ondo State field office and country office with the state epidemiologist designated as the incident manager. Disease surveillance and notification officers and volunteer contact tracing personnel recruited were trained by WHO Ondo State field office. Daily coordination meetings were held with situation reports of each strategic group presented and reviewed, and action points developed to strengthen ongoing response activities.

Surveillance and contact tracing

The Epidemiology and surveillance subgroup adopted the WHO standard case definition for suspected and confirmed LF cases and deaths to develop an operational case definition for case and death identification during active case search visits to health facilities and communities. The definition was any illness or death following a gradual onset and one or more of the following symptoms: malaise, fever, headache, sore throat, cough, nausea, vomiting, diarrhoea, myalgia, chest pain, hearing loss and a history of contact with excreta of rodents or with a case of LF from December 1, 2015 to April 30, 2016. Also, the name of case, age, sex, address, date of onset of illness, clinical symptoms of each suspected case or death identified were collected using the national LF case investigation form and Integrated Disease Surveillance and Response (IDSR) line-listing forms. Likewise, for any confirmed case or death of LF, contacts tracing and identification were carried out using the WHO definition for a contact, which was, “any person without any disease signs and symptoms but had physical contact which includes sharing the same room/bed, caring for a patient, touching body fluids, or closely participating in a burial with a case (alive or dead) or the body fluids of a case within the last three weeks from the onset of symptoms, within the hospital or community”. The identified contacts were enrolled into 21-day surveillance and follow-up exercise using the national viral haemorrhagic fever contact tracing line list and follow up guidelines and data tools. For each contact under surveillance, daily temperature monitoring and clinical evaluation were carried out for the period of follow up. Moreover, active case searches for additional cases (alive/dead) and retrospective review of hospital records were done in public and private hospitals in the LGAs.

Laboratory investigation and confirmation

Laboratory confirmation was performed at the Institute of Lassa Fever Research and Control, Irrua Specialist Teaching Hospital, Edo State, Nigeria, a national reference laboratory for LF diagnosis, treatment and research in Nigeria. The confirmation was based on a positive test using Lassa virus specific reverse-transcriptase PCR (RT-PCR). Laboratory samples (8mls of serum) were collected from each suspected case identified during the investigation. These are packaged in triplicate and transported in the reverse cold chain system by trained laboratory personnel to the laboratory. Feedbacks were provided to the state within 48-72 hours and in some instances more than 72 hours of receipt of specimens, due to unavailability of laboratory diagnostic materials.

Community/Social mobilization and health education

Information, Education and Communication (IEC) materials on LF were disseminated to the general public through various channels – print and electronic media, television adverts and radio, posters and banners (Appendix II). The main content of the messages was the discouragement of eating rats and poorly stored/rat-infested foods, symptoms of Lassa fever, and routes of transmission. Community and religious leaders were mobilized and empowered to inform and educate their members. Information on safe referral practices to the nearest hospital was also disseminated. Furthermore, the advocacy-communication subgroup in collaboration with the epidemiology and surveillance subgroup, trained health-care workers across hospitals in the state on case identification using the case definition, universal infectious disease precautionary measures in health care setting and on the importance of intensifying surveillance for additional cases.

Infection prevention & control

Procurement and distribution of infection and prevention control (IPC) supplies such as gloves, apron, alcohol, soap, chlorine, including complete Personal Protection Equipment (PPE) kit and IPC Standard Operation procedure (SOPs) guidelines to hospitals were carried out by the Infection prevention and control subgroup. Health workers exposed to confirmed LF cases were given 500mg of ribavirin on a 6 hourly basis for 5 days as PEP prescription. Decontamination exercises using chlorine and disinfectant kit were carried out daily in hospitals and vehicles used for transporting confirmed cases.

Case management

Suspected cases with severe clinical presentation while awaiting laboratory results and laboratory-confirmed cases were referred to Irrua Specialist Teaching Hospital in Edo State (about 158km away from Akure, the state capital). The hospital is one of the health institutions designated by the FMoH for LF case diagnosis, treatment and research in the country. Transportation of referred cases was carried out by the case management subgroup in collaboration with infection prevention and control subgroup in ambulances with drivers and personnel who have been trained on IPC.

Results

Ninety (90) suspected cases of LF from 8 LGAs (Owo, Akure South, Akoko North East, Akoko South East, Akoko South West, Akure North, Ondo West and Owo) of the 18 LGAs in the State were identified (Fig. 1). Of these cases, 19 were confirmed cases by the laboratory (Table 1). The confirmed cases were mainly from 3 LGAs (Akure South, Owo and Ose). The age group mostly affected among the confirmed cases were ≥ 15 years (73.7%) compared to the non-confirmed cases (85.9%), p=0.107) (Table 1). Case Fatality Rate (CFR) among the confirmed cases was 63.2% compared to 8.5% among non-confirmed cases (Table 1).

Fig. 1: Map showing the distribution of suspected Lassa fever cases by geographical location (LGAs) in Ondo State December, 2015 to April 30, 2016

Table 1: Characteristics of suspected Lassa fever cases in Ondo State December to April 30, 2016

Outcome of laboratory investigation
Variable Positive for Lassa fever Igm antigen, N=19 (21.1%) Negative for Lassa fever Igm antigen, N=71 (78.9%) p-Value
Age group (in years)
<5 4 (21.1) 4 (5.6) 0.107*
5-14 1 (5.3) 6 (8.5)
≥ 15 14 (73.7) 61 (85.9)
Mean age 29.3±1.9 33.6±1.8
Sex
Male 9 (47.4) 36 (50.7) 0.796
Female 10 (52.6) 35 (49.3)
Location (LGA)
Akure South 8 (42.1) 33 (46.5) 0.698*
Owo 10 (52.6) 28 (39.4)
Ose 1 (5.3) 2 (2.8)
Ondo West 0 (0.0) 3 (4.2)
Akure North 0 (0.0) 2 (2.8)
Akoko South East 0 (0.0) 1 (1.4)
Akoko South West 0 (0.0) 1 (1.4)
Akoko North East 0 (0.0) 1 (1.4)
Outcome of case management
Alive 7 (36.8) 65 (91.5) <0.001*+
Dead 12 (63.2) 6 (8.5)
Median days between onset of symptom and presentation at the 2.0 (0 -17) days 2.0 (0 -23) days
Exposure to a confirmed case
Yes 2 (10.5) 7 (10.0) 0.932
No/Unknown 17 (89.5) 64 (90.0)

*Fisher Exact Test, +Significant at p ≤ 0.05

Fig. 2 shows the epidemic curve of the outbreak, which began in the epidemiological week 53 of 2015 (28th December, 2015 to 3rd January 2016) with the index cases reported and confirmed in epidemiological week 1 of 2016 (4th to 10th January, 2016). Afterwards, there was a steady increase in the number of cases between epidemiological week 1 of 2016 (4th to 10th January, 2016) and epidemiological week 5 of 2016 (1st to 7th February 2016) with the outbreak reaching its peak in epidemiological week 4 (25th to 31st January, 2016) and week 5 (1st to 7th February, 2016) of 2016. Thereafter, fluctuations in the number of cases were noted with a steady decline in the cases recorded between epidemiological week 11 (14th to 10th March, 2016) and epidemiological week 14 (4th to 10th April, 2016) of 2016 (Fig. 2).

Epidemic curve of Lassa fever cases

Fig. 2: Epidemic curve of Lassa fever cases (suspected and confirmed) in Ondo State December 2015 to April 2016

Furthermore, 287 contacts of the confirmed cases were identified and followed-up as shown in Table 2 and Fig. 3. More than half of these contacts were females (64.5%) with most of them (89.2%) in the age group ≥ 25 years. Among the contacts, two (0.7%) developed symptoms that were consistent with LF (fever, malaise, sore throat, vomiting, diarrhoea) and were laboratory-confirmed. Of these two cases, one died while the other recovered after treatment (Fig. 4). Two hundred and seventy-six (96.2%) completed the follow-up exercise without developing any symptoms consistent with LF (Fig. 4).

Table 2: Characteristics of Lassa fever contacts followed-up from December 2015 to April 2016

Variable N (%) (n=287)
Age group (years)
<5 5 (1.7)
5-14 5 (1.7)
15-24 21 (7.3)
≥ 25 256 (89.2)
Gender
Male 102 (35.5)
Femalef 185 (64.5)
Occupation
Health care worker 241 (84.0)
Businessman/trader 30 (10.5)
Student 11 (3.8)
Teacher 4 (1.4)
Clergy 1 (0.3)
Type of contact
Hospital 240 (83.6)
Community 47 (16.4)
Neighbour 16 (5.6)
LGA
Akure South 144 (50.2)
Owo 125 (43.6)
Ose 14 (4.9)
Akure North 3 (1.0)
Idanre 1 (0.3)

Picture1

Fig. 3: Relationship of contacts with confirmed cases of Lassa fever in Ondo State, December 2015 to April 2016

outcome of lassa fever

Fig. 4: Outcome of Lassa Fever contact follow-up, December 2015 to April, 2016

Discussion

In this investigation, although a considerable number of suspected cases were identified, a few were laboratory confirmed indicating a highly sensitive surveillance system. In addition, more than half of the confirmed cases were female with the majority in the adolescent and the young adult age group. The CFR was high despite early detection and reporting of suspected cases indicating challenges in the case management protocol. Furthermore, contacts tracing and follow-up were successful, to some extent, with only very few contacts developing the disease.

The finding in which the proportion of those with disease was higher among females compared to males is inconsistent with a previous outcome of investigation of LF in Nigeria as reported by Ajayi et.al. 2013.13 Differences in exposures between male and female may have been responsible for this finding, given that gender role have not been shown to be a significant factor in the transmission of LF and other VHFs.23 Most of the confirmed LF cases in this investigation were in the age group ≥ 15 years. Ajayi et.al, 2013 in a similar investigation of LF outbreak in Ebonyi State Southeast Nigeria reported that almost all the cases recorded during the outbreak were in the older age group of ≥ 20 years.13

The high CFR (63.2%) among the laboratory-confirmed cases in this outbreak is slightly lower than those reported in other part of Africa such as Sierra Leone (69%), and however higher than those reported previously in several outbreaks in Nigeria.24 For instance, Getso et.al, 2014 reported a CFR of 40% during an outbreak of LF in Taraba state, northern Nigeria in the year 2012.25 Similarly, Ajayi et.al, 2013 reported a CFR of 40% during an outbreak of LF in Ebonyi state, south-eastern Nigeria.13 The higher CFR reported in our investigation could be attributed partly to the strategy of case management adopted by the case management subgroup where LF cases were referred outside the state for treatment due to the absence of a well equipped infectious disease treatment facility to manage these cases in the state. Furthermore, the quality of care received by the LF cases could have played some roles as the treatment cost at the referral hospital were partly the responsibility of the family members/caregivers of these cases thereby leaving families that were poor with inability to pay for good-quality health care for their wards. However, information on the cost of treatment for LF case and the economic burden on caregiver/families were not captured in our investigation.

The epidemic curve of this outbreak suggests a continuous source pattern of transmission, which is inconsistent with previous outbreaks reported in Nigeria.13 Apart from the two contacts who developed the disease, there are no known human sources of the disease for the others. Contrary to what we observed, previous investigations have shown that transmission of the VHFs such as LF has been related to direct contact with blood and other bodily fluids of people who are acutely ill.23

The spread of the disease during the outbreak was contained within a short period of time through an effective public health response strategy using the IMS/EOC model. This model uses an inter-sectoral collaborative approach in response to disease and other public health condition outbreaks resulting in coordination and resource mobilization. It relies substantially on surveillance for cases and contacts, case management and infection control, and laboratory diagnosis of cases, as well as effective public information and communication.

Previous studies on disease outbreak control have also reported the use of the EOC model in prompt and successful containment of diseases and other public health condition outbreaks in Nigeria and African regions.21,22,26 Shuaib et.al, 2014 reported that the use of the EOC model to coordinate the outbreak response and consolidate decision making during the Ebola Virus Disease outbreak in Nigeria in 2014 was largely significant with helping to contain the disease outbreak early in Nigeria.21

Similarly, Adeyanju et. al, 2015 in the investigation of the acute methanol poison outbreak in Ondo state, Nigeria, reported the use of the EOC model as an efficient outbreak response model in the coordination and response to outbreaks of public health conditions.22 Also, Kouadio et. al, 2016, reported that the use of the EOC model contributed to a high-level implementation of the National Polio Emergency Operational Plans strategies and activities in Nigeria with a resultant reduction in the number of WPV cases from 122 in 2012 to 53 (57% reduction) in 2013; and 6 (90% reduction) in 2014.26

The outbreak did not record any known nosocomial transmission of the disease among health-care workers who treated confirmed LF cases prior to their referral. Transmission among health-care workers has been a common occurrence in previous LF outbreaks in Nigeria with high incidence and fatality recorded among health workers providing care for confirmed LF and other VHFs cases.5,11,13,25,27 The high awareness of the disease among health-care workers, with a consequence of minimizing exposure may be partly responsible. The awareness might have created fear to care for LF patients, due to inadequate infra-structural and logistic support to the health facilities. Although several awareness creation and sensitization trainings on infection prevention and control in hospital settings were conducted for health workers across the state, these workers were not provided with PPEs materials and tools to protect themselves in most cases. There were no holding centres or isolation wards for VHF in any facility in the State. The infectious disease centre in the State capital, built to care for VHFs and other infectious diseases was not functional, due to lack of Manpower. The confirmed LF cases were transferred to the specialist hospital outside the State on receipt of the confirmation results, sometimes in an undesirable manner. Health workers who were highly exposed to confirmed LF cases were not provided with oral ribavirin as a post exposure prophylaxis on time, some of them were reported to procure the drugs on their own. We, however, did not obtain data on PEP to evaluate its impact.

Furthermore, the finding that most of the contacts of the LF confirmed cases identified, were females, is consistent with Iroezindu et. al, 2015 who reported similar findings during an outbreak of LF in Enugu southeastern Nigeria.28 The cultural practices in Nigeria, where the social burden of caring for the sick falls to the lot of women thereby increasing their vulnerability to infectious diseases may partly explain the observation. In addition, the high proportion of health- care workers identified as contacts for follow-up exercise is consistent with previous investigation,28 hence the need to provide adequate infection preventive measures within the health care setting to reduce exposure of health-care workers to infectious diseases.

Conclusion

Finally, the main limitation of this investigation was our inability to conduct a risk factor survey to identify factors associated with the cause of the outbreak. However, the strong collaboration among partners as well as the incident management approach to the outbreak was a key to the control and containment of the outbreak.

One key lesson learnt from the investigation was that the confirmed cases were mainly primary cases; hence the need to focus on measures of breaking the chain of transmission in the animal-man interphase in LFs epidemic preparedness and response. In addition, the high case fatality rate despite early reporting and investigation suggested the need for a review of the case management policy and structure in the State. There is the need to ensure that holding wards are established in all the health facilities and that trained, and highly motivated manpower are provided at these and the isolation centres at the State Capital and other specialist hospitals across the State.

Funding

The authors received no specific funding for this work.

Competing Interests

The authors have declared that no competing interests exist

Corresponding Author

Elvis E. Isere: Email: elvisisere@gmail.com; Phone number: +23408030480305

Data Availability

Data are available in Figshare using the following link: https://figshare.com/s/76a75e5a0292267576d3.

APPENDIX

Incident Management system structure used for the investigation of the outbreak and coordinating public health response

Appendix I: Incident Management system structure used for the investigation of the outbreak and coordinating public health response

IEC posters produced for sensitization during the outbreak by the SMoH and health development partner

Appendix II: IEC posters produced for sensitization during the outbreak by the SMoH and health development partner

Source of exposure of contacts to confirmed case of Lassa fever in Ondo State, December 2015 to April 2016.) 22

Appendix III: Source of exposure of contacts to confirmed case of Lassa fever in Ondo State, December 2015 to April 2016

]]>
http://currents.plos.org/outbreaks/article/lessons-learnt-from-epidemiological-investigation-of-lassa-fever-outbreak-in-a-southwest-state-of-nigeria-december-2015-to-april-2016/feed/ 0
Zika virus outbreak in Suriname, a report based on laboratory surveillance data http://currents.plos.org/outbreaks/article/obk-17-0011-zika-virus-outbreak-in-suriname-a-report-based-on-laboratory-surveillance-data/ http://currents.plos.org/outbreaks/article/obk-17-0011-zika-virus-outbreak-in-suriname-a-report-based-on-laboratory-surveillance-data/#respond Thu, 10 May 2018 10:00:07 +0000 http://currents.plos.org/outbreaks/?post_type=article&p=75436 Introduction : Since the identification of ZIKV in Brazil in May 2015, the virus has spread extensively throughout the Americas. Cases of ZIKV infection have been reported in Suriname since October 2, 2015.  Methods : A laboratory-based surveillance system was quickly implemented according to previous experience with the emergence of chikungunya. General practitioners and public health centers located in different districts of Suriname were asked to send blood samples from suspicious cases to Academic Hospital for molecular diagnosis of Zika virus infection. We investigated Zika-related laboratory data collected during surveillance and response activities to provide the first outbreak report in Suriname in terms of time, location and person. Results : A total of 791 molecularly confirmed cases were reported during a 48-week interval from October 2015 to August 2016. The majority of ZIKV-positive cases involved women between 20 and 39 years of age, reflecting concern about Zika infection during pregnancy. The outbreak peaked in mid-January and gradually spread from the district of Paramaribo to western coastal areas. Discussion : This report provides a simple and comprehensive description of the outbreak in Suriname and demonstrates the utility of laboratory data to highlight the spatiotemporal dynamics of the outbreak in that country.

]]>

Introduction

Zika virus (ZIKV) is a mosquito-borne flavivirus that is closely related to yellow fever and dengue viruses and can be transmitted by the bite of an infected Aedes aegypti mosquito, through sexual contact 123 or from mother to fetus 4 . The first large outbreaks, were not reported until 2007 from the Island of Yap in Micronesia and in May 20155, the World Health Organization reported the first local transmission of ZIKV in the north east of Brazil6. Since this initial detection, this virus has spread extensively throughout the Americas678910111213.

Although ZIKV infections have not historically been regarded as a significant public health concern, during this recent emergence, the virus has been linked to neurological disorders and severe congenital abnormalities. As at July 2017, 48 countries and territories have confirmed autochthonous, vector-borne transmission of ZIKV disease, while five countries have reported sexually transmitted Zika cases. Cases of ZIKV infection have been reported in Suriname since October 2, 2015; this nation, which has 530,000 inhabitants, was one of the first South American countries to report Zika virus infections, after Brazil and Colombia.

Phylogenetic analysis has indicated that the isolated virus belongs to the Asian genotype and appears to be most closely related to a strain that was circulating in French Polynesia in 2013.

This article describes the laboratory based surveillance system in Suriname and the incidence of confirmed cases of ZIKV infection according to sex, age, and spatial and temporal distribution

Methods

General practitioners and public health centers located in different districts of Suriname were asked to send blood samples from suspicious cases to Academic Hospital for molecular diagnosis of Zika virus infection. In particular, in cases involving pregnant women, samples were screened for free upon request. Viral RNA was extracted from blood and urine samples using a RNeasy Mini Kit (Qiagen, Hilden, Germany). An in-house molecular real-time RT-PCR assessment based on an approach detailed by Lanciotti et al.14 was used to confirm all cases. Records from patients with positive results between October 2, 2015, and August 23, 2016, were extracted into a database for epidemiological analyses. Collected data included the date of sample collection, which was used as a proxy for symptom onset; age; gender; and the patient’s district of residence, which was used as a proxy for location. Clinical cases reported by the Suriname Ministry of Health to Pan American Health Organization were compared to the confirmed cases. A clinical case of ZIKV disease was defined as a person with a rash with at least two of the following symptoms: fever higher than 38°C, conjunctivitis (non purulent/hyperemic), arthralgia, myalgia and peri-articular edema.

This analysis is based on data collected during the surveillance and response activities implemented during the ZIKV outbreak in Suriname. All data used in this report were aggregated so that they could not be associated with any specific individual.

Results

A total of 3,502 samples (2,752 blood samples and 750 urine samples) were collected from 3,460 individuals, and laboratory evidence of ZIKV infection was found in 791 cases. The outbreak spread rapidly throughout the country, reaching all 10 different districts in Suriname. The peak of the epidemic was observed in mid-January (W2016-03), with 107 molecularly confirmed cases (Figure 1). The weekly number of confirmed cases followed the same dynamic as clinical cases reported by the Ministry of Health.

Fig. 1: Temporal and spatial distribution of ZIKV cases in Suriname, October 2015-August 2016

Overall, 69.9% (553/791) of ZIKV-positive cases involved women (Table 1).

Table 1

Characteristics of 791 patients with molecular confirmation of Zika virus in Surinam, October 2015 – August 2016.

Characteristics N (%) Incidence per 100,000 population
Sex
Female 553 (69.9) 85.1
Male 238 (30.1) 201.7
Age group
<20 years old 111 (13.7) 55.6
[20-39] 413 (51.2) 242.9
[40-59] 190 (23.6) 160.6
>59 years old 93 (11.5) 147.4
District
Brokopondo 1 (0.2) 12.0
Commewijne 19 (3.3) 60.5
Coronie 5 (8.8) 147.4
Marowoijne 4 (0.7) 19.7
Nickerie 39 (6.9) 113.9
Para 6 (1.1) 24.3
Paramaribo 358 (63.0) 148.6
Saramacca 11 (1.9) 62.9
Sipaliwini 4 (0.7) 10.8
Wanica 121 (21.3) 102.3
Total in Surinam 791 147.7

The majority of ZIKV-positive cases involved patients between 20 and 39 years of age, who accounted for more than 51.2% of cases, reflecting concern about Zika infection during pregnancy that led to the overrepresentation of this population among tested individuals. Small fractions of cases involved patients between 0 and 19 years of age (13.7%) and patients older than 60 years of age (11.5%). More than 60% (358/791) of ZIKV-positive cases were from the district of Paramaribo, which was the first district to report confirmed cases, representing an incidence rate of 1.48 confirmed cases for every 1,000 inhabitants (vs 0.71 for every 1,000 inhabitants in other districts). From October 2, 2015, to November 11, 2015, 11 confirmed cases were reported, all of which were within this district. The outbreak gradually spread to western coastal areas and remained active for an extended period in the western districts of Nickerie and Coronie.

Discussion

A laboratory based surveillance system for ZIKV infections was quickly implemented in Suriname according to the previous experiences with chikungunya and has monitored ZIKV outbreak dynamics in the different territories of the country.

The Zika virus outbreak represents a major public health threat, particularly for fetuses of infected pregnant women. Even if the aforementioned statistics substantially underestimate the total impact of the outbreak in Suriname because they do not account for unreported clinical illnesses or asymptomatic infections, this report provides a simple and comprehensive description of the outbreak in Suriname and demonstrates the utility of Zika-related laboratory data to highlight the spatiotemporal dynamics of the outbreak in that country. Although the number of involved pregnant women included in the cohort is not available, follow-up of pregnant women who were infected during the outbreak will be critical for improving understanding regarding the spectrum of adverse pregnancy and infant outcomes associated with Zika virus infection and identifying the effects of certain factors, such as the timing of infection during pregnancy.

This report confirms that the timely and passive routine reporting of spatiotemporal information from clinical and laboratory data is critical for determining and communicating infection risks and for implementing risk reduction activities in high-risk areas, especially in the context of a new emerging infectious disease. To adapt prevention messages and activities and improve knowledge, it is essential to rely on a representative multisource surveillance system based on clinical and confirmed cases with particular attention to complications related to neurological disorders, congenital abnormalities and children born from infected mothers.

Competing Interest Statement

Dr. C. Flamand, on behalf of all the authors of the manuscript submitted to PLoS Current Outbreaks declare that no competing interests exist.

Data Availability Statement

Weekly laboratory data used in the article are accessible in Supplementary file S1, representing the minimal dataset publicly available. Since de-identified data in this report will not constitute truly anonymous information considering that in some situations (Date of collection, municipalities, sex, age), it could be possible to subsequently link the de-identified data back to an identifiable individual, access to individual data is restricted. Interested researchers may send requests to the head of laboratory of Academic Hospital in Paramaribo, Dr. John Codrington (johncodrington@hotmail.com) Laboratory Director Academic Hospital, Flustraat #1 P.O. Box 9305 Paramaribo, Suriname ; Tel : 011-597-442222

Supplementary file S1

Zika virus laboratory data, Academic Hospital-Paramaribo, October 2015 – August 2016.

Epi-week Confirmed cases Diagnosis tests
W2015-38 0 10
W2015-39 0 6
W2015-40 2 8
W2015-41 0 7
W2015-42 1 10
W2015-43 3 20
W2015-44 3 31
W2015-45 11 25
W2015-46 11 16
W2015-47 10 28
W2015-48 18 21
W2015-49 20 30
W2015-50 33 34
W2015-51 16 26
W2015-52 22 28
W2015-53 17 28
W2015-01 68 152
W2016-02 56 210
W2016-03 107 350
W2016-04 79 326
W2016-05 46 259
W2016-06 48 226
W2016-07 37 184
W2016-08 21 123
W2016-09 37 184
W2016-10 31 120
W2016-11 23 145
W2016-12 9 56
W2016-13 8 71
W2016-14 6 67
W2016-15 6 58
W2016-16 5 38
W2016-17 2 38
W2016-18 0 40
W2016-19 1 35
W2016-20 6 53
W2016-21 1 38
W2016-22 2 55
W2016-23 3 39
W2016-24 3 29
W2016-25 2 32
W2016-26 4 19
W2016-27 2 26
W2016-28 3 22
W2016-29 1 21
W2016-30 2 38
W2016-31 3 44
W2016-32 2 34

Corresponding Author

Claude Flamand: cflamand@pasteur-cayenne.fr

]]>
http://currents.plos.org/outbreaks/article/obk-17-0011-zika-virus-outbreak-in-suriname-a-report-based-on-laboratory-surveillance-data/feed/ 0